tabsFor() sorts by contribution priority when no user order is
set. Test expectations updated for sidebar defaultSize 400 and
decision ID D-1 (no zero-padding). PTY tests tagged forkpty and
run via dart test (forkpty output unreliable inside flutter test
runner). CI script adds --no-fatal-infos and --exclude-tags.
Co-Authored-By: Claude <noreply@anthropic.com>
Unused _resolvePtycPath (bin/clide.dart), _TrafficDot (app.dart
and clide_column_hat.dart), stale @override on _spawned field
(claude_pane.dart), duplicate import and unused local in
test_app.dart, unnecessary non-null assertions in project.dart.
Co-Authored-By: Claude <noreply@anthropic.com>
Platform-dispatched constants where Linux and macOS diverge:
TIOCSWINSZ, O_NONBLOCK, MsghdrDarwin struct (4-byte msg_iovlen
and msg_controllen vs Linux's 8-byte size_t fields), and split
recvmsg into Linux/Darwin typed variants so scm_rights uses the
correct struct per platform.
Also: app settings dir uses ~/Library/Application Support on
macOS, removed hardcoded TERMINFO from pane spawn env, removed
debug print from native_pty resize.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 40s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
macos-pty-problem.md: diagnosis of the SCM_RIGHTS fd issue (resolved
by switching to forkpty via NativePty).
pty-proposition.md: Gemini's diagnostic plan (cmsg_len analysis,
ptyc instrumentation, async recvFd).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
tmux determines window size from its client, not the PTY winsize.
TIOCSWINSZ + SIGWINCH on the master fd has no effect on tmux's
internal dimensions. Use tmux resize-window -t <session> -x -y
to explicitly set the window size on resize.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 27s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
tmux doesn't re-read the PTY winsize on SIGWINCH. After the ioctl
resize, call tmux refresh-client -C cols,rows to update tmux's
internal window dimensions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 27s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
tmux running inside a PTY without a traditional terminal client
defaults to a huge window size (2000+ cols, 10000 rows). Pass
explicit -x and -y flags matching the TerminalView dimensions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 34s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Reverts the bare-shell diagnostic. Spawns tmux new-session -A -s
<sessionName> for persistence. Claude command will be added back
after tmux + resize are stable.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Missed in the workspace → project rename. Consistent with the
project.open / project.validate / ProjectOpened naming.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
top fails with "Error opening terminal: xterm-256color" because the
PTY child can't find terminfo. Set TERMINFO=/usr/share/terminfo
explicitly in the pane environment.
Output flush changed from 16ms Timer to Duration.zero (next event
loop turn) for tighter batching of escape sequences.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 29s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Split escape sequences caused rendering artifacts (visible % prompt
mark, garbled lines between commands). Two fixes from the legacy
Python implementation:
1. Read buffer increased from 4096 to 65536 bytes. Larger reads mean
fewer chunk boundaries that can split multi-byte escape sequences.
2. Output writes throttled to ~60fps via a 16ms Timer. Multiple PTY
reads within one frame window are batched into a single
terminal.write() call, so the xterm parser sees complete sequences
instead of fragments that render as garbage between frames.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 33s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Firing all tier ticks simultaneously on project open causes multiple
IPC calls (pql, git, files) to run in parallel, their responses
triggering rapid setState rebuilds that beachball the merged thread.
Stagger 500ms apart so each panel refreshes sequentially.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Rapid resize during window drag sends many SIGWINCH signals causing
the shell to redraw repeatedly, corrupting xterm's render. Debounce
to 150ms so only the final size is sent.
NativePty.resize now explicitly sends SIGWINCH (signal 28) after
TIOCSWINSZ, matching the legacy Python implementation.
Removed 8px padding from ClidePtyView to eliminate potential
dimension mismatch between widget size and PTY cols/rows.
Default window size increased to 1600x900.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 1m57s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The PTY was spawning at 80x24 defaults before TerminalView had laid
out, then the real resize caused visual artifacts (ghost lines, broken
reflow). Now spawn triggers on the first onResize callback when real
dimensions are known — matching the pty-spike's proven pattern.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
execve doesn't search PATH — bare names like 'tmux' or 'claude' fail
silently. Resolve via the environment's PATH in Dart before passing
to the child. This matches ptyc's old execvp behavior.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 3m13s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
NativePty calls forkpty() directly — no helper binary, no socketpair,
no SCM_RIGHTS. The master fd stays in-process. Reader isolate uses
poll() for clean shutdown.
Based on the pty-spike proof-of-concept. Platform-aware: macOS uses
libSystem (DynamicLibrary.process), Linux needs libutil.so.1.
TIOCSWINSZ platform-detected.
PaneRegistry updated to use NativePty. registerPaneCommands no longer
needs a Toolchain parameter. All ptyc references removed from the
daemon layer.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
pql 1.4.4+ restores the 'decisions read' subcommand which returns
metadata + body + refs in one call. Reverts the temporary 'show
--with-refs' workaround.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Remove debug print statements. Pass --with-refs to pql decisions show
so cross-references render in the detail view.
The decision body text is not available from pql — it indexes metadata
only. The body section of the D-record lives in the markdown file.
Rendering it requires either a pql --with-context flag or direct file
read + section extraction.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The pql CLI command is 'decisions show', not 'decisions read'.
The client was sending the wrong subcommand.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The context panel was hidden or collapsed when a decision or ticket
was selected. activateTab alone doesn't make the panel visible.
Now both extensions explicitly setVisible + setCollapsed before
activating the detail tab.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Gitea remote kept as secondary (`gitea`). Updated repository URLs
in pubspec.yaml, licenses.yaml, and initial-plan.md. README rewritten
to reflect the current single-process Flutter architecture.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The IsolateClient and KernelServices now share the same DaemonBus.
Previously, backend events (pane.output, git.changed) went to a
separate bus that widgets couldn't see.
ClaudePane._spawnWhenReady waits for ProjectOpened before sending
pane.spawn, preventing "No project active" errors.
Recents loaded before runApp so the welcome screen shows them.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
macOS cmsghdr uses socklen_t (4 bytes) for cmsg_len, not size_t
(8 bytes like Linux). The Dart FFI struct was reading at wrong offsets
causing SCM_RIGHTS fd transfer to fail. Added CmsghdrDarwin with
Uint32 cmsg_len alongside the existing Linux layout.
SOL_SOCKET is 0xffff on macOS (was hardcoded to 1 for Linux).
Now platform-detected.
Switched from ad-hoc to Apple Development signing with team ID.
Removed app sandbox from entitlements — proper signing makes it
unnecessary. All SBPL hacks eliminated.
PtySession.spawn() runs recvFd in a child isolate to avoid blocking
the backend isolate's event loop.
Terminal testmode test now passes: pane.spawn via backend isolate
successfully forks and execs the shell.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 28s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
PtySession.spawn() now runs recvFd in a child isolate via
Isolate.spawn so the blocking FFI call doesn't stall the backend
isolate's event loop.
Terminal testmode test spawns a real backend isolate, opens a project,
and sends pane.spawn via IPC — the exact same path the full app uses.
Currently fails: ptyc successfully starts but its fork+execvp is
blocked by the macOS sandbox ("Operation not permitted"). The SBPL
allows /bin/zsh exec from the app process, but ptyc's child process
may not inherit the exec permission, or the PTYC_SOCK_FD is not
inherited by the ptyc child (Dart Process.start fd inheritance on
macOS).
IsolateClient.events getter exposed for test access.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 31s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
PtySession.spawn() uses synchronous FFI calls (socketpair, recvFd via
SCM_RIGHTS) that block the merged UI/platform thread on macOS. The
test correctly identifies this — PTY must move to the backend isolate.
Three tests added (echo, shell, write-to-child) that will run on
Linux where threads are separate. macOS skips with a diagnostic.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sends a valid JSON request to ptyc via stdin, verifies it parses the
request (exit 2 = syscall failed, not exit 1 = bad request). Full PTY
setup needs socketpair/SCM_RIGHTS which only PtySession provides.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 30s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Dropdown width 380 → 480 to prevent RenderFlex overflow on recent
project rows. "Close Workspace" → "Close Project". Keyboard shortcuts
show ⌘ on macOS instead of Ctrl.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Backend isolate now boots in two phases: resolve toolchain on spawn
(binary availability only), initialize services on project.open. The
dispatcher stays empty until a project activates — IPC requests before
that return "No project active".
Scheduler ticker only runs while a project is open. Fires an immediate
first cycle on ProjectOpened so sidebar panels refresh without waiting
for the next interval. Stops on ProjectClosed.
Renamed workspace → project throughout backend messages
(project.validate, project.open, project.ready), callbacks
(onProjectOpen, onValidateProject), and methods
(openProject, validateProject, resolveProject).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The merged UI/platform thread on macOS (Flutter 3.41) freezes on any
synchronous work — file I/O, Process.run, even isolate spawning during
early frames. No timing workaround (Timer, addPostFrameCallback,
Future.delayed) was reliable.
Fix: spawn a backend isolate that owns the DaemonDispatcher, GitClient,
PqlClient, FilesService, EditorRegistry, and Toolchain resolution. The
main isolate stays free for rendering. Communication uses SendPort with
the existing IPC message protocol (IpcRequest/IpcResponse/IpcEvent) —
zero new serialization.
New files:
backend.dart — spawns isolate, manages SendPort/ReceivePort
backend_entry.dart — isolate entry point, boots all services
isolate_client.dart — replaces InProcessClient for production
Toolchain now exposes resolveToolchainPaths() as a top-level function
with self-contained PATH expansion (no module-level state that would
prevent isolate message passing).
ClideTestApp gains boot-sequence tests: compute(), Isolate.run(),
sequential Process.run, and the full resolve+exec chain.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Teaches Claude when and how to use make run-testmode, which
category to run for a given change, how to interpret output,
and the pattern for adding new test cases.
Co-Authored-By: Claude <noreply@anthropic.com>
Six improvements to the testmode harness:
1. Shell passthrough tests use the resolved shell instead of
hardcoded /bin/zsh.
2. Exit code reflects test results (non-zero on any failure).
3. JSON summary line for machine-readable parsing.
4. IPC round-trip smoke (ping, version, unknown-cmd, encode/decode).
5. Extension lifecycle smoke (register + activate files, diff,
git, terminal; theme loading for all four bundled themes).
6. Test categories via CLIDE_TESTMODE dart-define (toolchain, ipc,
extensions, all). Makefile exposes TESTMODE_CATEGORY variable.
Also switches main.dart from bool.fromEnvironment to
String.fromEnvironment so category values other than "true" work.
Co-Authored-By: Claude <noreply@anthropic.com>
Replaces the transparent-background icons in the macOS asset
catalog with the black-circle variants from assets/logo/. The
64px size is downscaled from the 128px source since the master
set doesn't include that size.
Co-Authored-By: Claude <noreply@anthropic.com>
The package name is clide — the binary, bundle identifier, and
GTK application ID should match. Also aligns the macOS test host
paths and bundle identifiers.
Co-Authored-By: Claude <noreply@anthropic.com>
All six test/build scripts assumed a nested app/ package that
doesn't exist — the Flutter package lives at the repo root.
Removed every `cd app` and phantom path, consolidated the
redundant dart/flutter analyze passes in test.sh, and scoped
test_core.sh to core subsystem dirs so it no longer overlaps
with the full flutter test run.
Co-Authored-By: Claude <noreply@anthropic.com>
The hooks under .pql/hooks/ are installed locally by pql init —
they're runtime scaffolding, not source. The original block at
lines 49-51 already covers .pql/* with the pql-plan.json
negation; the second block was redundant.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 29s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The project switcher dropdown's "Open Local Project" action now uses
the native directory picker (same as the welcome screen). Shows "No
git repo found" dialog when the selected directory is not a git
repository.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
DebugProfile.entitlements gitignored (machine-specific sandbox paths).
native/dugite/ gitignored (downloaded at build time).
Linux runner: add pickDirectory handler using GtkFileChooserDialog
in the existing clide/window method channel.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
macOS sandbox blocks Homebrew git execution (symlinks resolve to Cellar
paths that SBPL cannot match). Bundled dugite-native — the same
relocatable Git distribution GitHub Desktop ships.
make dugite-fetch downloads the platform tarball to native/dugite/.
Toolchain checks the bundled binary first, sets GIT_EXEC_PATH and
GIT_TEMPLATE_DIR automatically. native/dugite/ is gitignored (57 MB
compressed).
Decision D-59 in tooling.md, dependency entry in licenses.yaml
(GPL-2.0 for the git binary, MIT for dugite-native build scripts).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Standalone test harness activated via CLIDE_TESTMODE=true dart-define.
Runs a table of existence checks and exec tests against all external
binaries (git, pql, tmux, ptyc, zsh), including zsh-passthrough and
env variants. Results printed to stdout and shown in a minimal UI.
Auto-exits after 15 seconds.
make run-testmode launches it with a 60-second kill timer and pipes
output to /tmp/clide-testmode.log.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Toolchain centralizes binary resolution — replaces five ad-hoc
mechanisms (expandedPath, _resolveGit, _resolve, _resolvePtyc,
_existsOnPath). Resolves via Future.delayed after runApp to avoid
blocking the merged UI/platform thread on macOS.
GitClient wraps all git operations with a typed API. Every subprocess
call goes through _run() using toolchain.git + toolchain.gitEnv.
Replaces free functions in operations.dart.
Native directory picker: NSOpenPanel on macOS (method channel in
AppDelegate), GtkFileChooserDialog on Linux. Falls back to text-input
dialog on web or MissingPluginException. Shows "No git repo found"
dialog when the selected directory is not a git repository.
PqlClient and pane commands updated to use Toolchain. ToolCheck
replaced by Toolchain.missing/allOk. All IPC handlers now catch
GitException to prevent unhandled exceptions on the merged thread.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
OS-detecting Makefile (make run works on macOS/Linux/Windows), 1280x720
default window in MainMenu.xib, squared app icons, sandbox with scoped
SBPL exceptions for subprocess execution, _DARWIN_C_SOURCE for ptyc
compilation, expanded PATH merging Homebrew and ~/.local/bin for GUI
apps, native traffic dots skipped on macOS (titlebar owns them).
DebugProfile.entitlements is gitignored (machine-specific SBPL paths);
a template with __HOMEDIR__/__PROJECTS__/__SHELL__ placeholders is
committed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 26s
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / integration_test (xvfb) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Pql skill: ticket show flags updated to reflect --with-context
replacing --with-decision and --with-children.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 26s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Replaces the N+1 parent-chain walk (one IPC call per ancestor, one
per decision ref) with a single pql ticket show --with-context call.
pql now returns ancestors and decisions arrays inline.
Updates PqlClient.ticketShow to accept withContext/withChildren flags,
removes the obsolete withDecision flag.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 35s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Theme-ui skill: document Phosphor Icons section with codepoint CSV
reference, icon lookup workflow, and bold/fill weight usage.
Pql skill: update ticket status command to reflect batch ID support
and removed state machine enforcement.
Changelog: scheduler service, auto-refresh, ClideAccordion, status
buttons, per-status ticket sections, codepoint CSV.
Co-Authored-By: Claude <noreply@anthropic.com>
Decisions: file-change subscription for decisions/*.md plus 1-minute
scheduler tick. Tickets: 1-minute scheduler tick plus changed-event
subscription from status button clicks. PQL markdown tab: file-change
subscription for *.md files. All panels get a manual refresh button.
Accordion sections use pin/focus logic: manually toggled sections are
pinned and stay open; the focused item's section auto-opens; unpinned
sections without focus auto-collapse. Both decisions and tickets views
use ClideAccordion with this pattern.
Ticket sidebar split into six status sections (IN PROGRESS, REVIEW,
READY, BACKLOG, DONE, CANCELLED) matching pql's actual statuses.
Status changes scroll the ticket into its new section.
Co-Authored-By: Claude <noreply@anthropic.com>
Status buttons in the detail view now call pql.tickets.status IPC,
publish a changed event to refresh the sidebar list, and reload the
detail. Disabled for the current status. IPC handler accepts ids as
a list for batch transitions. PqlClient.ticketSetStatus takes
List<String> and joins with comma per pql CLI contract.
Co-Authored-By: Claude <noreply@anthropic.com>
ClideAccordion — shared accordion section with optional leading widget,
replaces private _AccordionSection in tickets and decisions views.
Phosphor arrowClockwise icon (0xe036) for refresh buttons. Full
codepoint reference CSV (1512 glyphs) at assets/fonts/phosphor/ for
icon lookup without external tools.
Co-Authored-By: Claude <noreply@anthropic.com>
Tiered Timer.periodic instances (1min, 10min, 15min, 1hr, midnight)
running on a dedicated Dart isolate. Tick messages sent back to the
main isolate via SendPort and emitted as SchedulerTick events on the
DaemonBus. Extensions subscribe with events.on<SchedulerTick>() and
filter by tier.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 33s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Search tab defaults to ranked text search via pql search — debounced
300ms, scored results with visual score bar. DSL toggle switches to
raw PQL query mode for SQL-like queries. Clicking a search result
opens it in the markdown context viewer.
Adds pql.search IPC command and PqlClient.search method. Controller
gains SearchMode enum and search() method alongside existing runQuery.
Co-Authored-By: Claude <noreply@anthropic.com>
Search (PQL DSL query) is now the default left tab; Markdown (filtered
to .md files) on the right. Clicking a markdown file publishes on
the message bus; the markdown extension activates the context panel
viewer and re-publishes on a load channel after the frame so the
viewer is guaranteed to be mounted. Viewer subscribes to load channel
and editor buffer events, publishes focus for sidebar highlighting.
Co-Authored-By: Claude <noreply@anthropic.com>
Register files.read command for reading file content by relative path.
Remove top padding on context panel container and mainAxisSize.min
from ClidePaneChrome so the header sits flush and the pane fills
available height. Context panel max raised from 420px to 1000px.
Co-Authored-By: Claude <noreply@anthropic.com>
Each ClideCodeBlock and EditorView created its own TreeSitterService,
each allocating a WASM engine, store, parser, and cursor via FFI.
When widgets were disposed, multiple instances freeing the same
underlying native resources caused double-free corruption on startup.
Single shared instance for the app lifetime. Also fixes overlapping
tree-sitter spans that caused duplicated text in code blocks — spans
are now clipped to avoid re-emitting already-rendered characters.
Co-Authored-By: Claude <noreply@anthropic.com>
All sidebar and detail panes use clideFontSmall/clideFontBadge/
clideFontCaption instead of hardcoded sizes — single place to tune.
Decision detail view subscribes to the message bus for selection
events (same pattern as ticket controller) and publishes focus after
loading. Both decisions and tickets sidebar views subscribe to the
focus channel, expand the accordion section, highlight the active
card, and scroll it into view.
Clickable DQRT links wired in decision detail, ticket detail, and
markdown viewer via onRecordTap. Ticket descriptions now render
through ClideMarkdown instead of plain text.
Co-Authored-By: Claude <noreply@anthropic.com>
Ctrl+Plus/Minus/0 text zoom (5% steps, 60%–200%) via MediaQuery
textScaler. App-wide clideLineHeight (1.25) applied in DefaultTextStyle.
Context panel drag resize was inverted — dragging left shrank the
panel instead of growing it. Negate delta for right-anchored slots.
Default sidebar widened from 240px to 400px, context panel from 280px
to 420px (both at their previous maximums).
Co-Authored-By: Claude <noreply@anthropic.com>
Inline grouping collapses consecutive inline nodes (strong, em, code,
a, del) into a single RichText widget instead of rendering each as a
separate block — fixes tight list items like `builtin.git` appearing
on its own line. HTML entity unescaping moved from pre-parse (useless)
to post-AST extraction where the markdown package actually introduces
the entities. Explicit Josefin Sans Light font family and 1.25 line
height on all RichText spans. Body text bumped to 16px; inline code
uses clideFontMono for visual weight parity.
Adds onRecordTap callback — markdown links matching [DQRT]-\d+ render
as tappable WidgetSpans with hover underline. New typography constants
clideFontSmall (12), clideFontBadge (11), clideLineHeight (1.25).
Co-Authored-By: Claude <noreply@anthropic.com>
More vertical breathing room between paragraphs (8→12), list
items (2→6), headings (increased top/bottom). Tighter horizontal
indent on lists (16→8) and blockquotes (12→8). Added HTML entity
unescaping (quot, amp, lt, gt, apos) before parsing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both tickets and decisions extensions now uncontribute and
re-contribute the detail tab on every selection, passing the
selected ID via initialId. This guarantees the view loads
immediately — no first-click miss from the widget not being
built when the message fires.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New pql.decisions.read IPC verb returns the full markdown section
body. Decision detail view now shows: title card, full decision
text (rationale, cost, cross-references), and clickable ref cards
for cross-reference navigation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Clicking a parent ticket in the tree publishes
builtin.tickets/selection — loads that ticket in the detail view.
Clicking a decision ref publishes builtin.decisions/selection —
switches context panel to the decision detail. Both show hover
highlight. Enables full cross-reference navigation without
leaving the context panel.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Detail controller publishes builtin.tickets/focus with the loaded
ticket ID after every load. Ticket list subscribes and highlights
the matching card with selected background and accent border.
Works regardless of what triggered the detail load — sidebar
click, parent tree navigation, or future command palette.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extension listens to its own selection channel and activates the
detail tab in the context panel — fixes the chicken-and-egg where
the controller only subscribes after the widget builds. Registration
order in main.dart now determines sidebar icon rail order: tickets
first, then decisions, files, git, pql, problems.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Icons now declared on TabContribution via the icon field — the
hardcoded switch in _BottomRail is a fallback only. All sidebar
and context extensions set their Phosphor icon explicitly.
PanelRegistry supports setTabOrder(slot, List<String>) — tabs
render in the order specified by project.layout.sidebar.order /
context.order in settings.yaml. Falls back to registration order.
Priority field removed from all contributions.
Context tabs are static again — persist with their last content,
selection just loads new data and switches focus. No spawn/despawn.
Tickets sidebar priority set to -200 (leftmost, default open).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two flow fixes:
1. First click now works — extension passes initialId to the
detail view constructor, which loads immediately in
didChangeDependencies. No second click needed.
2. Tab despawns when user navigates away — extension listens to
PanelRegistry changes, despawns detail tab when it's no longer
the active context panel tab.
Each selection re-creates the tab with a fresh initialId so the
view always matches the selection. Applied to both tickets and
decisions.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Context panel tabs are no longer static. Extensions spawn them
dynamically when relevant and despawn on deactivate:
- Tickets: detail tab spawns on builtin.tickets/selection
- Decisions: detail tab spawns on builtin.decisions/selection
- Backlinks: spawns when editor opens a file
- Markdown viewer: spawns when editor opens a .md file
- Graph: removed from static context (will be command-spawnable)
The icon rail only shows what's active — no empty tabs sitting
there waiting for content.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Click a decision card in the sidebar → publishes on
builtin.decisions/selection → detail view loads full record →
context panel activates decisions.detail tab and renders:
- Type dot with tooltip, ID colored by type, domain badge
- Title, date
- Status badge for resolved questions
Same MessageBus pattern as ticket detail.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Click a ticket card in the sidebar → publishes on
builtin.tickets/selection → detail controller loads full ticket,
walks parent chain, resolves decision refs → context panel
activates tickets.detail tab and renders:
- Full ticket header with type dot, ID, priority
- Status controls (tappable row to change status)
- Description
- Parent tree (indented compact cards)
- Referenced D-records as cards
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
DaemonBus (was EventBus): typed events for system/IPC layer.
MessageBus: channel-based pub/sub for UI/extension coordination.
Messages carry publisher (auto-stamped from extension ID),
channel (required), timestamp, and payload. Subscribe by
publisher, channel, or both — zero collision across extensions.
Extension context gains publish() and subscribe() convenience
methods that auto-stamp the extension's ID as publisher.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both views now group items into collapsible accordion sections:
- Decisions: CONFIRMED/QUESTIONS/REJECTED with colored dots
- Tickets: ACTIVE/BACKLOG/DONE/OTHER with caret toggle
Cards show type dot with tooltip, ID in mono, domain badge
(decisions), parent reference (tickets), wrapping title, and
status badges. Search filter forces all sections open. ACTIVE
and BACKLOG default expanded; DONE default collapsed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
TicketTypeColors maps initiative/epic/story/task/bug to distinct
colors, with dark and light presets auto-selected from the active
theme. Ticket IDs are now color-coded by type (purple initiative,
blue epic, green story, muted task, red bug). Status dot remains
for state. Pattern documented in theme-ui skill as the standard
for extension-owned domain colors.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
pql 1.2.0 changed its record format to drop leading zeros. All
262 references across 18 files renumbered (D-001→D-1, T-043→T-43,
etc.). pql-plan.json export updated as the git source of truth
for the planning database. DB rebuilt via pql plan import.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both views were calling pql.exec (not registered). Switched to
pql.decisions.list and pql.tickets.list which are registered
IPC handlers. Response parsing updated to read structured data
instead of parsing JSON from stdout.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Decisions and tickets are now top-level sidebar sections with
their own extensions and filter boxes. The pql panel keeps only
Files and Query — its own surfaces. Removed loadDecisions,
loadTickets, _DecisionRow, _TicketColumn from pql controller
and view.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ClideTappable builder now receives (context, hovered, pressed).
Optional onLongPress and onPressChanged callbacks. ClideButton
is now a StatelessWidget wrapping ClideTappable. ClideIconRail
refactored — _RailButton uses ClideTappable with tooltip.
ClideTooltip positions smart (above when near screen bottom).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Encapsulates the hover + click + cursor pattern repeated across
the codebase. Builder receives (context, hovered) so callers
control their own hover styling. Refactored _ThemeLink to use it.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Status line aligned to the right. Theme name is clickable —
highlights on hover and executes theme.pick command to open
the theme picker dialog.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ToolCheck kernel service checks ptyc/pql/tmux/git availability
once at boot, shared by status bar and welcome screen. Welcome
status line shows "application ok" in green or lists missing
tools in amber. Logo doubled to 144px. DialogHost moved inside
Expanded to fix blank workspace. Welcome maxWidth 850px, path
overflow fixed with Flexible.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Hat bar moved above the Stack so it stays visible when the welcome
overlay shows. Recent project path row uses Flexible with ellipsis
to prevent horizontal overflow.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Click "clide > project" label to open a dropdown with:
- Search/filter box for recent projects
- Recent projects list with path, branch, and time ago
- Open Local Project action (path dialog)
- New Window (spawns detached process)
- Close Workspace (returns to welcome screen)
Dropdown uses DialogRouter, dismissible via Escape or backdrop
click. Reuses ProjectManager.recents and ClideFilterBox.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three format engines adopted per the new D-058 rule (format
engines clear the dependency bar; UI chrome stays in-house):
- jovial_svg 1.1.26 (BSD-3) — SVG renderer via CustomPaint
- markdown 7.2.2 (BSD-3) — GFM parser; renderer is ours
- flutter_widget_from_html_core 0.17.2 (MIT) — HTML renderer
Welcome screen now loads logo.svg via jovial_svg instead of a
raster PNG. POLICY.md updated with format engine rule. All three
documented in licenses.yaml.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Two clean sets from their SVG sources:
logo.svg → logo-{16..1024}.png (bare, transparent)
appicon.svg → appicon-{16..1024}.png (on dark circle)
All rendered from 2048px masters. References updated in
pubspec.yaml, welcome view, and native runner.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
SVG rendered at 2400 DPI to 2048px master, then downsampled to
1024/512/256/192/128/48/32/16. Crisper edges at all sizes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Logo SVGs made square (160x160). New onbackground variant has the
clide logo on a #20202C filled circle. Rendered at 1024px and
downsampled to 256/128/48 for crisp taskbar icons. Native runner
loads clide-taskbar-256.png instead of the bare logo.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The relative path failed because flutter run's cwd differs from
the bundle directory. Now reads /proc/self/exe to find the binary
location and builds the icon path from there.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Flutter's globalPosition is window-relative, not screen-absolute.
Now using gdk_device_get_position() on the native side to get
the actual screen coordinates at the moment startDrag/startResize
is called. Dart no longer passes coordinates — the native handler
queries the pointer directly.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The KDE server-decoration protocol negotiation compiles and runs
but KWin still applies SSD. XWayland remains the only reliable
path on KDE Wayland + GTK3. The native protocol code stays in
place for when KWin or GTK behavior changes. Q-031 tracks this.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The xdg-decoration protocol files caused a linker error
(xdg_toplevel_interface undefined) since we switched to the KDE
server-decoration protocol. Removed. Added make build-check
target that runs flutter build linux to catch native compilation
failures that flutter test misses.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prebuilt ptyc binary sits alongside libtree-sitter.so in the
platform-specific native directory. Resolution order:
native/linux-x64/ptyc → ptyc/bin/ptyc → ~/.local/bin/ptyc → PATH.
Eliminates the make ptyc-build step on fresh clones.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace xdg-decoration (needs xdg_toplevel, GTK3 doesn't expose it)
with org_kde_kwin_server_decoration (works with wl_surface which
GDK provides). On KDE Wayland: binds the manager from the registry,
creates a per-surface decoration object, requests mode=None (0).
Falls back gracefully on GNOME (protocol not present, GTK's
set_decorated(FALSE) already works). X11 fallback via
gdk_window_set_decorations(0) unchanged.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace XWayland fallback with proper xdg-decoration protocol
negotiation. On Wayland: binds zxdg_decoration_manager_v1 from
the compositor registry, requests CLIENT_SIDE mode on the toplevel
surface — KWin respects this and drops server-side decorations.
Falls back to gdk_window_set_decorations(0) + set_decorated(FALSE)
on X11. GDK_BACKEND=x11 removed from Makefile.
Also fixes mouse-jumping glitch: startDrag and startResize now
pass screen coordinates from Flutter's globalPosition through the
method channel to gtk_window_begin_move/resize_drag.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Without OS decorations there are no resize handles. Added 6px
edge and 12px corner invisible hit zones around the window
perimeter via ClideResizeBorder. Each zone changes the cursor
and calls gtk_window_begin_resize_drag via the clide/window
method channel with a GdkWindowEdge direction.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Documents the workaround (GDK_BACKEND=x11), rollback instructions,
the proper fix (libwayland-client + xdg-decoration protocol), and
the alternative timeline (Flutter GTK4 migration).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
GTK3 doesn't implement the xdg-decoration Wayland protocol, so
KWin always adds server-side decorations regardless of
gtk_window_set_decorated(FALSE). GDK_BACKEND=x11 forces XWayland
where MOTIF_WM_HINTS reliably removes decorations. Revisit when
Flutter moves to GTK4 or implements xdg-decoration natively.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
gtk_window_set_decorated alone doesn't work on KDE Wayland — KWin
adds server-side decorations regardless. Now also calling
gdk_window_set_decorations(gdk_win, 0) after realize and before
first frame show, which communicates directly with the compositor
via GDK's Wayland layer.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
KWin adds server-side decorations that ignore
gtk_window_set_decorated(FALSE). GTK_CSD=1 forces client-side
decoration mode, which our set_decorated(FALSE) then suppresses.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New tokens: chromeBackground, chromeForeground, chromeBorder —
shared root for hat bar, sidebar, context panel, status bar,
spines, and drag handles. All resolve to bgSunken/textDim/border
in the palette by default. Themes can override individually to
diverge surfaces. Replaces direct sidebarBackground/statusBar
references in chrome contexts. Updated theme-ui skill.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Guides widget development to use the correct background, border,
text, and hover tokens per surface type. Documents the three-tier
background model (root/chrome/surface), the token identity rule
(no borrowing across surfaces), and flags the pending need for
chromeBackground/chromeForeground tokens.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The non-line pixels in the 5px drag handle now fill with the
adjacent panel's background color (sidebar or context) instead
of being transparent. The 1px divider line stays on the inner
edge touching the center column.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sidebar pads left/top only (not right — that's the divider side).
Context panel pads right/top only (not left — divider side). The
divider handle's transparent pixels now show the center column's
background correctly.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The 1px visible line in the 5px drag handle now sits on the
innermost pixel (touching the center column) instead of centered.
Side panels (sidebar and context) get 2px padding on all sides
except the bottom, creating a balanced inset that matches the
divider's visual weight.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Handle is 5px wide for comfortable drag, with a 1px center line
using dividerColor (switches to panelActiveBorder on hover).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Hat bar is now a single continuous row above the three columns
instead of per-column segments. Left: macOS traffic lights or
empty. Center: project > branch. Right: minimize/maximize/close
on Linux or empty. Column separators keep 4px hit area but show
a 1px centered line.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Hide GTK title bar via gtk_window_set_decorated(FALSE). Add
MethodChannel('clide/window') for drag/minimize/maximize/close
wired to GTK window functions. Dart WindowControls service wraps
the channel. Three per-column hats in RootLayout: left (macOS
traffic lights or plain drag), center (project > branch label),
right (minimize/maximize/close glyph buttons on Linux). Resolves
Q-006.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Shared filter box with search icon, debounced input, and clear
button. Applied to all six left-panel panes for consistent
filtering: Files (flat path results when filtering), Git (filter
staged/unstaged/untracked by path), Decisions (by ID/title/domain),
Tickets (by ID/title/status), Problems (by source/message), and
pql Query (replaces custom _QueryInput).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Shows "● ok" in green when both ptyc and pql are available.
When either is missing, shows each broken tool by name in amber
("● ptyc not found", "● pql not found"). Font matches the git
branch widget — uses inherited UI face at clideFontCaption size,
no mono override. Dot indicator replaces the Phosphor icon.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 35s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
The null byte in the --format string didn't survive Process.run
argument passing, so the output was unparseable and no branches
were returned. Switched to pipe separator with lastIndexOf split
to handle branch names containing pipes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Branch picker now handles Escape key to dismiss, shows an × close
button in the header, and displays error/empty states instead of
silently showing nothing. Dialog backdrop is click-to-dismiss (the
dialog content stops propagation so clicks inside don't close it).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The tool status indicator now checks the repo-local ptyc binary
before falling back to PATH lookup, matching the resolution logic
in main.dart.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Flutter app now looks for ptyc/bin/ptyc relative to the
workspace root before falling back to PATH. Fixes
"ProcessException: No such file or directory. Command: ptyc"
when running in-process without ptyc installed globally.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
InProcessClient dispatches IPC requests directly to handlers in the
same process, bypassing the unix socket entirely. All subsystem
registries (pane, files, editor, git, pql) are created in the
Flutter app's boot sequence and registered on a DaemonDispatcher.
Events bridge to the kernel EventBus via _BusEventSink. The app
no longer needs `clide --daemon` running.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
pubspec.yaml is now the single source of truth for version and
project metadata. Makefile reads version from pubspec.yaml. All
references to project.yaml across CLAUDE.md, CHANGELOG.md,
decisions, and skills updated.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The status bar now shows whether ptyc and pql are findable on PATH
(green = found, amber = missing) instead of the obsolete daemon
connection state. Per D-056 there is no daemon to connect to.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Rewritten to reflect the dissolved daemon, single Flutter package
at repo root, and updated repo layout. Removed references to app/,
clide --daemon, two-package structure, and AOT compilation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
These files are generated and updated by pql init, not manually
authored. Tracking them causes drift when pql versions change.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Contains last-opened project, panel sizes, collapse state — all
personal preferences. Committing it causes merge conflicts when
users have different layouts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Single Flutter package at the repo root. All code, tests, assets,
and platform directories moved from app/ to root. Package renamed
from clide_app to clide — all imports rewritten. Merged pubspec
combines core (ffi) and app (flutter, yaml, xterm) dependencies.
Makefile simplified: no APP_PRESENT conditionals, no cd, no daemon
lifecycle. 317 tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
D-055: multiple Claude sessions share the workspace as internal
tabs inside the Claude pane header. Primary has no close; secondaries
show ×. Tab row hidden when only primary exists. + button and
double-click-empty spawn new secondaries.
Session names now use a readable path slug (strip $HOME, replace /
with -) with fallback to FNV-1a hash when the slug exceeds 80 chars.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Side panel content was vertically centered by Container's default
alignment. Added Alignment.topLeft so file tree, viewer, and other
panel content docks to the top.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Status bar text (git branch, connection indicator) used
clideFontCaption at 12px — too small relative to the new 15px app
default. Bumped the full type scale: body 14→15, caption 12→13,
mono 13→14.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phosphor glyph fontSize set to 1.0 in unit space so glyphs fill
the full icon size (16px in the rail). Welcome tests updated to
match the redesigned view.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phosphor icons: reduced font size from 0.88 to 0.75 in unit space
and center-aligned paragraph to prevent horizontal stretching.
Icon rail: removed top border, tightened padding (10h/4v from
12h/6v), reduced icon size to 14px to fit the 26px strip. Status
bar: added horizontal padding and vertical center alignment.
Bottom strip row: added top border, crossAxisAlignment.stretch
so all three sections fill the height uniformly. Default app font
bumped from 14 to 15px.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Icon rails pulled out of _SidebarSlot and _ContextSlot into the
bottom row alongside StatusbarHost. All three strips — left icon
rail, app strip, right icon rail — now sit on one continuous
horizontal line per D-047. Sidebar and context panels render
content only; their bottom rails share the status bar row with
matched widths. Collapsed spines get a matching-width spacer.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The daemon now checks for ptyc at ../ptyc/bin/ptyc relative to its
own binary location before falling back to PATH lookup. Fixes
"ProcessException: No such file or directory. Command: ptyc" when
ptyc is built but not on PATH.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Welcome content block is now centered in the viewport with a
720px max-width constraint. Status line pinned to the bottom.
Font sizes bumped across the board: title 52px, subtitle 16px,
section headers 12px, action labels 15px, recent project names
15px. Debug banner disabled via debugShowCheckedModeBanner.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Welcome screen is now a full-window overlay that covers the IDE
when no project is open. Two-column layout: START actions (open
folder, clone from git, start Claude session) with keyboard
shortcut hints, and RECENT projects list showing name, path,
branch, and relative timestamp. Status line shows version, daemon
connection state, and active theme.
ProjectManager tracks up to 10 recent projects with path, branch
name, and last-opened time, persisted as JSON in app.recentProjects.
Boot chain: try last project → try cwd → show welcome. Clicking a
recent project opens it directly into Claude.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ProjectManager.open() now saves app.lastProject to user settings.
Boot calls openLast() instead of opening cwd — first launch shows
the welcome screen, subsequent launches restore the last project
directly into Claude. The welcome "Open project" button shows a
modal path-entry dialog that validates against git and activates
the Claude tab on success. Removed the dead "New Claude session"
button since secondary sessions are spawned via command palette.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
On desktop boot, main.dart calls project.open(cwd) to resolve the
git root. If a project opens successfully, the workspace activates
claude.primary instead of welcome. The _WorkspaceSlot now listens
to ProjectManager: when isOpen, Claude is always the visible
primary surface; when no project, welcome shows. This closes the
gap where the app loaded the project root but still displayed the
welcome screen as if no choice had been made.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The delayed show callback fired even after the mouse had already
exited, causing tooltips to stack. Track hover state with a bool
and check it after the delay before showing.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Four stub extensions now contribute real tabs with views:
- Decisions panel (sidebar, priority -20): lists confirmed D-records
from pql decisions list with ID, title, and domain.
- Tickets panel (sidebar, priority -10): lists tickets with status
dot color-coded by state (done=green, in_progress=blue,
cancelled=red).
- Markdown viewer (context, priority -100): shows raw content of
the active .md file, listens for editor.buffer_activated events.
- Graph view (context, priority -80): lists files with inbound and
outbound link counts from pql search --connections.
All four fetch data via the daemon's pql.exec IPC surface.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phosphor Icons v2.0.8 (MIT) — regular, bold, and fill weights
bundled under assets/fonts/phosphor/. PhosphorIconPainter bridges
the font glyphs into the existing ClideIconPainter interface via
Paragraph API so icon rails, pane headers, and any widget using
ClideIcon can consume them. Sidebar and context panel icon rails
now use Phosphor glyphs instead of hand-painted CustomPaint icons.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Collapse state, panel sizes, active sidebar/context tabs, and
editor split ratio save to .clide/settings.yaml on every layout
change and restore on boot. Guards against missing projectDir
so tests and no-project scenarios degrade gracefully.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Editor opens above Claude via Ctrl+E with a draggable divider at
35% height (clamped 15-70%). Ctrl+W or Escape closes it. The
workspace no longer renders a tab bar — Claude is the always-visible
primary surface per D-047/D-048; the editor is a split overlay
per D-049, not a tab. The _WorkspaceSlot separates editor from
primary content tabs and renders them as a vertical split when the
editor is open.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Collapsed side panels render as a 12px vertical spine with rotated
label, hover highlight, and badge dot (D-051). Focus mode snapshots
and restores the full layout arrangement (D-052). Fourteen new
commands in the default-layout extension: collapse toggles
(Ctrl+Shift+1/3), panel focus (Ctrl+1/2/3), focus mode (Ctrl+.),
sidebar section switching (Alt+1-5), Escape dismiss (D-054). Right
panel gets a bottom icon rail matching the sidebar pattern (D-047).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
POLICY.md codifies runtime behavior rules (no network on default
launch path), dependency vetting, vendored binary rebuild process,
telemetry prohibition, and licensing requirements. Binding on all
contributors including agents.
Line length 160 in .editorconfig (all files) and
analysis_options.yaml (dart formatter).
Co-Authored-By: Claude <noreply@anthropic.com>
Vendor libtree-sitter.so (v0.26.8, wasmtime 44.0 statically linked)
and call its C API through dart:ffi. Grammar WASM files are loaded by
tree-sitter's embedded wasmtime engine via ts_wasm_store_load_language.
48 grammars compiled with tree-sitter build --wasm, 48 highlight
queries (42 from upstream repos, 6 written in-house for comment,
erlang, gdscript, gitignore, hcl, hlsl, swift). SQLite grammar
replaces full SQL (974K-line parser exhausted RAM at compile time).
Removes wasm_run and wasm_run_flutter — they downloaded a 22 MB
binary from GitHub at first launch, violating the no-network-on-
default-launch-path policy.
Co-Authored-By: Claude <noreply@anthropic.com>
Core frame builtins are infrastructure the shell can't function
without. Content extensions (editor, claude, canvas, etc.) are
bundled but architecturally removable — users swap the surface,
not the data.
Removes builtin.jira stub: Jira belongs as a third-party
extension, not a frame builtin. Git is a split component —
daemon-side process stays in frame, UI surfaces are extensions.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 41s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Larger wordmark (48px), START section header, icon-prefixed
action rows for Open project and New Claude session. Semantics
hint preserved for a11y.
Co-Authored-By: Claude <noreply@anthropic.com>
The discard button in the git panel now opens a dialog warning
that changes will be permanently lost. Prevents accidental data
loss from a misclick.
Co-Authored-By: Claude <noreply@anthropic.com>
git.branches and git.checkout IPC verbs. Status bar shows current
branch with ahead/behind count; clicking opens a branch picker
dialog for checkout.
Co-Authored-By: Claude <noreply@anthropic.com>
ClidePalette renders as a floating overlay with fuzzy-filtered
command list. Ctrl+Shift+P toggles it. Ctrl+B toggles sidebar,
Ctrl+J toggles context panel. All wired through CommandRegistry
via the default layout extension.
Co-Authored-By: Claude <noreply@anthropic.com>
Four new icon painters (git branch, search, terminal, warning).
ClideIconRail widget with parent-owned hover state so only one
icon highlights at a time. Sidebar slot renders content above +
icon rail below; workspace and context slots keep the tab bar.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 40s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
PID file at /tmp/clide-daemon-<uid>.pid replaces pkill pattern
matching. make run depends on stop so stale daemons are killed
before build. Flutter exit via ctrl-C no longer aborts Make.
Co-Authored-By: Claude <noreply@anthropic.com>
Surface map now tries palette keys (bgSunken, surfaceHi, textHi,
etc.) first, falling back to semantic refs for legacy themes.
Sidebar/panel bg differentiates from workspace bg. Buttons use
accent colour with onAccent text. Four-level text gradient flows
to widgets.
WCAG-AA contrast fixes: clide textDim lightened, midnight
onAccent darkened, paper textDim darkened. Contrast test updated
to cover all four themes. Resolver tests updated for the new
palette shape. Goldens regenerated.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 40s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Kill stale daemon before build to prevent "Text file busy". Trap
EXIT + INT + TERM for reliable cleanup on ctrl-C. Add make stop
for standalone daemon kill.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 39s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
SurfaceTokens gains seven syntax-role fields (keyword, type,
string, number, comment, method, punct). Theme loader parses
the syntax: YAML block and injects values into the surface
override layer. Resolver defaults map syntax roles to semantic
colours so themes without explicit syntax blocks still compile.
Four new theme YAMLs (clide, midnight, paper, terminal) replace
summer-night. Palette keys mapped to the existing semantic layer;
syntax values pass through directly. Goldens regenerated for the
14→font-size change.
Co-Authored-By: Claude <noreply@anthropic.com>
New tokens/ directory with YAML palette + syntax files for all
four themes, plus a pure dart:ui Dart file (no Material). README
updated to reflect the no-Material constraint and open typography
question.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 41s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
D-043: adopt token palettes from design, reject Material wrapper.
D-044: four bundled themes replace summer-night.
D-045: syntax highlighting tokens in the theme pipeline.
Q-025: body text face — mono everywhere vs Josefin Sans + mono.
R-012: MaterialApp wrapper rejected (conflicts with D-007).
Co-Authored-By: Claude <noreply@anthropic.com>
Four self-contained HTML mockups (wireframe, flows, hi-fi, design
system), four Dart theme files (clide, midnight, paper, terminal),
and flat PNG renders. Reference material for the UI build — not
runtime assets.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 41s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Replace Flutter default icons across Linux (GTK window icon via
bundled asset), macOS (AppIcon.appiconset all sizes), and web
(favicon, PWA icons, manifest, meta description). Logo PNGs at
16–1024px generated from the SVG source. Window title set to
"clide" everywhere. Web manifest colours match summer-night.
Co-Authored-By: Claude <noreply@anthropic.com>
Builds the daemon binary if stale, starts it in the background,
launches flutter run -d linux, and cleans up the daemon on exit.
Co-Authored-By: Claude <noreply@anthropic.com>
Sidebar and panel text was too small — dozens of bare fontSize:
10/11/12 values fighting the ambient DefaultTextStyle. Root font
bumped from 13 to 14. Three semantic constants replace the magic
numbers: clideFontBody (14), clideFontCaption (12), clideFontMono
(13). Widgets that render body text now inherit instead of
overriding.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 40s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
PqlExtension contributes a sidebar tab (files, query, decisions,
tickets views) and a context-panel tab (backlinks + outlinks for
the active file, auto-refreshing on editor.active-changed).
ProblemsExtension contributes a sidebar tab aggregating pql.doctor
and decisions.sync diagnostics with actionable hints.
Both upgraded from stubs to 0.1.0.
Co-Authored-By: Claude <noreply@anthropic.com>
pql init --with-skill=yes installs .claude/skills/pql/SKILL.md
covering vault queries and the planning surface (decisions +
tickets). Lock file tracks version + SHA-256 for drift detection.
Bash(pql) and Bash(pql *) added to .claude/settings.json so
pql commands run without permission prompts.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 37s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
GitController hydrates from git.status IPC and auto-refreshes on
git.changed events. Panel shows staged/unstaged/untracked/conflict
groups with per-file actions and an inline commit field.
DiffController renders unified diffs with line numbers, addition/
removal colouring, and a staged/unstaged toggle. Both extensions
upgraded from stubs to 0.1.0.
ClideText gains an optional fontFamily parameter so diff lines can
render in JetBrainsMono without breaking golden tests (font still
inherits from ambient DefaultTextStyle by default).
Co-Authored-By: Claude <noreply@anthropic.com>
Maps clide git status|diff|stage|unstage|discard|commit|log|stash|
pull|push to the git.* IPC verbs. Updates --help to list the new
tier-3 subcommands.
Co-Authored-By: Claude <noreply@anthropic.com>
Shell-outs to git for status (porcelain v1/v2), unified-diff
parsing, and operations (stage, unstage, hunk-apply, discard,
commit, stash, log, pull, push). IPC verbs git.* registered on
the daemon dispatcher with git.changed event emission on
mutations. 42 new core tests.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 38s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
EditorController mirrors the daemon's active buffer into the Flutter
UI. Hydrates on mount (editor.active → editor.read), subscribes to
editor.opened / active-changed / edited / saved / closed, and pushes
local edits back via editor.set-content. A one-shot echo-suppression
flag keeps the daemon's authoritative edit echo from clobbering the
caret the user just moved.
Text surface is EditableText directly — no TextField / Material
import, so D-007's "bare WidgetsApp" guardrail carries into the
editor tab. JetBrainsMono via the shared mono-family constants;
cursor + selection colours come from the theme tokens.
File-tree click now calls ipc.request('editor.open', {path}) instead
of the kernel command registry (which wouldn't have reached the
daemon). The opened file becomes active, the editor tab's controller
sees editor.active-changed, and the text content reconciles.
Linux release bundle still builds; 174 app tests + 74 core tests
stay green.
Co-Authored-By: Claude <noreply@anthropic.com>
`bin/clide` gains the single-word shortcuts CLAUDE.md's tier 2 spells
out: open, active, insert, replace-selection, save. Each maps the
flat positional argv into the canonical editor.* IPC shape. Insert
and replace-selection accept a lone `-` to read text from stdin so
piping works (`pbpaste | clide replace-selection -`).
`clide tail --events` is the subscribe mode. Same socket as the
request side; the client just reads + filters events. --filter
SUBSYSTEM or SUBSYSTEM:ID narrows the stream. Exits cleanly on
SIGINT.
defaultSocketPath() now respects CLIDE_SOCKET_PATH before XDG — the
existing override callers always had this up their sleeve (via
XDG_RUNTIME_DIR manipulation) but making it explicit unblocks
parallel test runs where each test needs its own daemon socket. The
new end-to-end CLI suite does exactly that: 5 tests spin up real
daemon subprocesses and exercise the shortcut surface through the
live IPC stack.
74 core tests pass; round-trip verified by hand (open README.md →
insert → tail --events captures editor.opened / edited /
selection-changed / saved).
Co-Authored-By: Claude <noreply@anthropic.com>
EditorBuffer + Selection + EditorRegistry hold the daemon-side
active-file model (D-006 subsystem 'editor'). Active buffer
tracking means `clide insert "…"` and `clide replace-selection
"…"` target the UI's focused file without the caller supplying an
id. Mutations mark buffers dirty; editor.save writes back to disk
through the workspace root; events fire on every state change so
subscribers can mirror.
IPC surface matches CLAUDE.md's tier-2 list + the natural extras
(list, read, activate, set-selection, set-content, close). Tests
cover open-idempotence, insert at caret, replace-selection range
swap, dirty→save→clean round-trip, close picks a new active
buffer, out-of-range selection clamping.
69 core tests pass.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 37s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Pane is now a pure data class — id, kind, pid, argv, cwd, title,
isClosed. The daemon-side PaneRegistry holds a parallel map of
PtySession keyed on id; registry methods look up both sides when
writing / resizing / closing.
The `clide.dart` barrel no longer re-exports `src/pty/*`,
`src/panes/registry.dart`, or the `*_commands.dart` modules — all
three transitively import `dart:ffi` which isn't available when
compiling to WebAssembly. The daemon entrypoint (bin/clide.dart) +
core tests import them via deep paths now. Pane / PaneKind /
DaemonEventSink / RecordingEventSink stay in the barrel since
they're pure data the Flutter app references over IPC.
Verified: `dart analyze` clean, 53 core tests green, 174 app tests
green, `make ui-smoke` compiles + serves + Playwright smoke passes,
daemon boots + ping round-trips + SIGTERMs cleanly.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Failing after 45s
test / integration_test (xvfb) (push) Has been skipped
test / bundle smoke (xvfb 5s) (push) Has been skipped
test / daemon subprocess + web WASM smoke (push) Has been skipped
Tier-0 stub flipped to the real Claude pane. Primary-per-repo
semantics: session name is `clide-claude-<hash>` where <hash> is
FNV-1a over the workspace root path, so reopening clide attaches
to the same `claude` process through `tmux new-session -A`.
Primary has no close button — closing the tab disposes the widget
but deliberately does NOT call pane.close, so the tmux session
survives until the user actually exits claude or clide is shut
down. Secondary sessions (spawned via the claude.new-secondary
command registered here, UI wiring lands next) close normally and
the pane.close cascades into tmux kill-session.
Graceful fallback when tmux isn't on PATH: retries the spawn with
argv=['claude'], surfaces "no-tmux · fresh every launch" in the
header subtitle so the user knows persistence is off.
Session-naming unit tests cover determinism + uniqueness per repo.
Full app suite: 174 tests passing.
Co-Authored-By: Claude <noreply@anthropic.com>
Captures the session-persistence + primary-vs-secondary policy
referenced from ptyc/README.md and deferred through the Tier 1
substrate work. Primary per repo (keyed on git root) survives app
restart via `tmux new-session -A`; secondaries are ephemeral and
collapse back to the primary on close; primary has no close
affordance. Daemon owns the session lifecycle, the UI just asks
pane.spawn / pane.close and observes events.
Required by the next commit (builtin.claude) which consumes this
policy.
Co-Authored-By: Claude <noreply@anthropic.com>
Flutter sidebar tab that lazy-loads the workspace tree via IPC
files.ls and refreshes subtrees on files.changed events. Click-to-
open routes through a future editor.open command; until Tier 2
registers it, the execute call no-ops gracefully.
Daemon side adds a new files subsystem:
- files.root returns the resolved workspace root (git root if
present, otherwise cwd)
- files.ls lists a directory with ignore filtering applied
- files.watch starts a recursive Directory.watch and fans
FileSystemEvents out as files.changed IPC events
- FilesService owns the watcher + ignore set lifecycle
IgnoreSet + IgnorePattern implement the common gitignore subset:
anchored (/foo), directory-only (foo/), negation (!foo), **
crossing dirs, ** at trailing position. Built-in layer hides clide-
owned dirs (.git, .pql, .clide, .dart_tool, build, node_modules);
.gitignore + .clideignore at the root layer on top per D-004. Full
multi-file ignore_files: layering from .pql/config.yaml is future
work.
11 new ignore-matcher tests + 5 files.* dispatcher tests.
Co-Authored-By: Claude <noreply@anthropic.com>
Flips the Tier-0 stub into a real extension. Contributes a Terminal
tab in the workspace slot that spawns $SHELL -l via IPC pane.spawn,
feeds pane.output events into xterm.dart's Terminal, and routes user
input back through pane.write. Viewport resize propagates via
pane.resize. Surface for disconnected-daemon / exited-shell states so
there's no silent dead tab.
Knows nothing about Claude deliberately — the Claude-specific pane
(primary-per-repo, tmux-backed, D-041) lives under builtin.claude in
the next steps.
Co-Authored-By: Claude <noreply@anthropic.com>
ClidePtyView is a theme-bound wrapper around xterm.dart's TerminalView
— token-derived TerminalTheme, JetBrainsMono as the face, Semantics
live-region label so screen readers + Playwright both hear it. The
consumer (terminal / Claude extensions) owns the `Terminal` model and
wires IPC pane.write / pane.output → terminal.write() themselves; the
widget deliberately has no IPC dependency so it stays trivially
testable.
ClidePaneChrome is the shared pane header — title + subtitle + leading
icon + trailing widgets + optional close button. The close button is
null-conditional so primary Claude panes (D-041, landing in step 7)
can render without one.
xterm 4.0.0 added as a justified runtime dep + logged in
licenses.yaml per D-042. 3 new widget tests cover header rendering,
close-button presence, and the close-tap round-trip.
Q-023 records the SSH-remote-development question so the daemon + IPC
seams don't accrete local-only assumptions during Tier 1-5.
Co-Authored-By: Claude <noreply@anthropic.com>
Implements the Tier-1 pane subsystem from D-006: spawn / list / focus /
close / write / resize / tail commands, plus pane.spawned / output /
exit / resized / focused / closed events. PaneRegistry owns per-pane
PtySession lifecycles and id generation (p_N); a DaemonEventSink seam
keeps pane code decoupled from the IPC server package.
DaemonServer.broadcast() fans events out to every connected client.
Per-client subsystem/id filtering (`tail --filter pane:p_7`) is
deferred — Tier 1 broadcasts everything and the subscriber discards.
Panes carry a `kind:` field (terminal | claude). Step 7 (builtin.claude)
adds the claude-specific pane flow on top of this generic substrate —
the subsystem itself stays neutral.
14 new core tests: registry unit coverage (spawn → pane.spawned event,
output → base64 events, write/resize/close round-trips, idempotent
close, claude kind on the wire) plus dispatcher coverage (argv
validation, unknown-id → not-found, text vs bytes_b64, etc). All 37
core tests pass in ~3s under test-core.
Co-Authored-By: Claude <noreply@anthropic.com>
PtySession wraps the ptyc helper: socketpair + Process.start + recvmsg
with SCM_RIGHTS for master-fd transfer, a background isolate that
loops on blocking read() and posts byte chunks, plus write/resize/
kill/close. close() SIGTERMs the child so the PTY's EOF wakes the
reader naturally; SIGKILL + fd close + isolate kill cover the edge
where the shell ignores SIGTERM — avoids the known Linux quirk where
closing an fd doesn't unblock an in-flight read() on it.
Env defaults stamp TERM=xterm-256color, COLORTERM=truecolor,
CLICOLOR_FORCE=1 so shells + tmux + Claude emit 24-bit sequences
that xterm.dart can render. User env (HOME / USER / SHELL) still
inherits via mergePtyEnv().
ffi: 2.1.3 added as a runtime dep — the FFI bindings for socketpair,
recvmsg, read/write, and ioctl(TIOCSWINSZ) need an allocator we're
not writing by hand. Justified in pubspec + listed in licenses.yaml
per D-042.
make test-core (ci/test_core.sh) runs the Flutter-free core tests
under a 120s hard timeout with setsid + process-group kill, wired
ahead of the fast app tests in push-check so a hung PTY test can't
wedge a pre-push. Current core suite: 24 tests in ~1s.
Co-Authored-By: Claude <noreply@anthropic.com>
test / unit + widget + golden + a11y (push) Successful in 1m30s
test / integration_test (xvfb) (push) Failing after 1m14s
test / bundle smoke (xvfb 5s) (push) Failing after 1m3s
test / daemon subprocess + web WASM smoke (push) Successful in 1m42s
Three sections instead of one flat list. `self:` carries clide's own
MIT license (rendered first in the About screen so the user knows
what they're running before the dependency list). `dependencies:`
covers artefacts that actually ship in the binary — JetBrainsMono,
JosefinSans, yaml. `dev_dependencies:` tracks build-time tooling
(mocktail, alchemist, flutter_lints, lints, test) for audit
completeness without polluting the user-facing About panel.
Also backfilled the two root-package dev deps that were missed the
first pass: `lints: 5.0.0` and `test: 1.25.8` (root pubspec, not
app/). Root LICENSE mirrored into app/assets/LICENSE + declared as
a bundled asset so the About screen can read it at runtime (Flutter
can't reference paths above the package root).
D-042 updated to describe the three-section split and the dev-vs-
runtime distinction.
Co-Authored-By: Claude <noreply@anthropic.com>
Every third-party artefact shipping in the clide binary — fonts,
Dart packages, native tools, bundled assets — now has an entry in
app/assets/licenses.yaml with name, kind, version, homepage, license
identifier, relative path to the bundled license text, and a
one-line purpose. The About screen (Tier 6) will render this file
verbatim. Seeded with the current set: JetBrainsMono, JosefinSans,
yaml, mocktail, alchemist, flutter_lints.
D-042 captures the two-step-commit rule (artefact + licenses.yaml
entry in the same changeset) alongside D-031's prefer-zero-deps
budget: preferring zero deps is the budget; licenses.yaml is the
visible consequence when the budget grows.
CLAUDE.md "Dependencies & supply chain" gains a matching guardrail
line. The per-dep license text files (OFL.txt for both fonts) are
declared as pubspec assets too so the About screen has something to
display, not just a manifest.
Co-Authored-By: Claude <noreply@anthropic.com>
Vendor the JosefinSans variable font (upright + italic, weight range
100-700), OFL licensed. _AppRoot installs it as the ambient
DefaultTextStyle at weight w300 (Light); ClideText inherits rather
than re-specifying the family, so Alchemist's Ahem injection keeps
goldens deterministic per D-024 while production renders the
intended face.
typography.dart gains clideUiFamily + clideUiFamilyFallback
alongside the existing mono constants; .gitignore learns about
Alchemist's failures/ diff output dir so local re-runs don't leak
into `git status`.
Co-Authored-By: Claude <noreply@anthropic.com>
Vendor the 4 core weights (Regular, Italic, Bold, BoldItalic) of
JetBrains Mono v2.304 under app/assets/fonts/jetbrains_mono/, ~1 MB
total. OFL license checked in alongside. Declared as the
JetBrainsMono family in pubspec so Flutter bundles it automatically.
app/lib/widgets/src/typography.dart exposes clideMonoFamily plus a
platform-ordered fallback chain (SF Mono / Menlo / Monaco →
JetBrains Mono system install / Fira Code / Hack / DejaVu / Liberation →
Cascadia / Consolas → monospace) for web builds and test harnesses
that don't load asset fonts. Terminal widgets pick both up via a
single import.
Co-Authored-By: Claude <noreply@anthropic.com>
One-shot helper that opens a PTY, forks a child, execvp's the given
argv with the slave as stdin/stdout/stderr, and hands the master fd
back to the caller over a unix socket via SCM_RIGHTS. Wire contract
is stdin JSON → stdout JSON + fd transfer per D-005; socket fd
defaults to 3 with PTYC_SOCK_FD override for language runtimes whose
subprocess machinery shuffles the low fd numbers.
Minimal JSON parser (no deps) scoped to the exact accepted shape.
Exec-failure pipe (CLOEXEC) reports child-side errors back without
zombies. Window size applied via TIOCSWINSZ before fork; child
becomes session leader + makes the slave its controlling TTY.
Root Makefile PTYX_PRESENT typo fixed → PTYC_PRESENT, and a
ptyc-test target added alongside ptyc-build / ptyc-clean. Eight
smoke tests pass (happy path, env replacement, cwd, window size,
bad argv, type errors, exec failure, unknown keys).
Co-Authored-By: Claude <noreply@anthropic.com>
2026-04-21 22:27:20 +02:00
606 changed files with 38109 additions and 4389 deletions
@@ -63,10 +63,10 @@ Cutting a release is its own commit. In a single commit:
1. Move all entries from `## [Unreleased]` under a new heading `## [X.Y.Z] — YYYY-MM-DD`.
2. Leave an empty `## [Unreleased]` section at the top with its subsection skeleton ready.
3. Bump `project.yaml``version:` to `X.Y.Z` (drop the `-dev` suffix for the tag; re-add it on the next development commit if desired).
3. Bump `pubspec.yaml``version:` to `X.Y.Z` (drop the `-dev` suffix for the tag; re-add it on the next development commit if desired).
4. Commit subject: `release vX.Y.Z`.
`project.yaml` is the single source of truth for the version — the Makefile reads it for ldflag stamping of the sidecar binary, and the Flutter app reads it for build info. Bumping `project.yaml` and the changelog out of sync is the mistake this rule prevents.
`pubspec.yaml` is the single source of truth for the version — the Makefile reads it for ldflag stamping of the sidecar binary, and the Flutter app reads it for build info. Bumping `pubspec.yaml` and the changelog out of sync is the mistake this rule prevents.
-`docs/ADRs/` directory — content lifted into `decisions/` as D/R
@@ -79,7 +559,7 @@ heading, and (b) bumping `project.yaml` `version:` in the same commit.
- Go sidecar skeleton under `sidecar/` — `cmd/clide/main.go`, `go.mod`, and the `internal/*` packages (`cli`, `daemon`, `diag`, `git`, `ipc`, `pql`, `proc`, `pty`, `version`). Deleted wholesale per [ADR 0005](docs/ADRs/0005-dart-core-ptyc-peer.md): the "sidecar language: Go" premise no longer holds once the core is Dart. All functionality listed for those packages will be reimplemented under `lib/` as part of Tier 0.
- Go-specific Makefile targets (`lint`, `vuln`, `test-race`, `fmt`, `tidy`, `snapshot`, `tools`, `install` via Go), the `govulncheck`/`goimports`/`golangci-lint` version pins, and the pre-push hook's `GOBIN` PATH injection. Replaced with Dart/Flutter equivalents (`analyze`, `format`, `test`, `test-integration`, `build` via `dart compile exe`).
-`module:` and `go_version:` from `project.yaml` — single-language core means no Go module path to track.
-`module:` and `go_version:` from `pubspec.yaml` — single-language core means no Go module path to track.
### Changed
@@ -125,14 +605,14 @@ heading, and (b) bumping `project.yaml` `version:` in the same commit.
[ADR 0004](docs/ADRs/0004-ignore-file-strategy.md) — Ignore file strategy (`ignore_files:` in `.pql/config.yaml`, layered).
- Pre-push quality gate: `.githooks/pre-push` runs `make push-check` (lint + test + test-race + test-integration + vuln + app-analyze + app-test) so bad pushes are caught locally before they hit Gitea. The app-side targets gracefully noop until Flutter is scaffolded. Install with `make hooks` (sets `git config core.hooksPath .githooks`); the hook prepends `$GOBIN`/`$HOME/go/bin` to PATH so govulncheck resolves without the user touching their shell profile.
- Go sidecar/CLI skeleton under `sidecar/` (module `git.schweitz.net/jpmschweitzer/clide/sidecar`): `cmd/clide/main.go`, `internal/cli` with a stdlib-flag dispatch, `internal/diag` mirroring pql's exit-code + stderr-JSON contract, `internal/version` with ldflag-stamped build info, and placeholder packages for `daemon`, `pty`, `proc`, `git`, `ipc`, `pql` awaiting their tier. `clide --version` emits JSON build-info today.
- Root `Makefile` drives both the Go sidecar and the Flutter app under one toolchain. Version is read from `project.yaml` via awk and stamped into the sidecar via `-ldflags -X`. Flutter targets gracefully noop before the app is scaffolded so the Makefile is usable from day one. Pinned Go tooling (govulncheck, goimports, golangci-lint) installs via `make tools`.
- Root `Makefile` drives both the Go sidecar and the Flutter app under one toolchain. Version is read from `pubspec.yaml` via awk and stamped into the sidecar via `-ldflags -X`. Flutter targets gracefully noop before the app is scaffolded so the Makefile is usable from day one. Pinned Go tooling (govulncheck, goimports, golangci-lint) installs via `make tools`.
-`ci/` entry scripts: `test.sh`, `lint.sh` (includes the supply-chain gate — no green lint without a green CVE scan), `security.sh`, `release.sh` (stub).
- Project identity files for the Flutter rebuild at the repo root: `project.yaml` (single source of truth for version + module path, version 2.0.0-dev), a fresh `README.md`, MIT `LICENSE`, and `.editorconfig`. The Python clide's manifest and README are preserved under `legacy/`.
- Project identity files for the Flutter rebuild at the repo root: `pubspec.yaml` (single source of truth for version + module path, version 2.0.0-dev), a fresh `README.md`, MIT `LICENSE`, and `.editorconfig`. The Python clide's manifest and README are preserved under `legacy/`.
- [`docs/initial-plan.md`](docs/initial-plan.md) — the north-star design document for the Flutter rebuild. Captures what we kept from Python Clide (pane model, git skills, Claude-always-visible), what we took from Obsidian (canvas and graph — no vault, no bases, no plugin inheritance), what Claudian's short experiment contributed (Go sidecar, CLI-first, pql-as-subsystem, ignore-file strategy), and the tier roadmap (Tier 0 app+sidecar handshake → Tier 5 canvas+graph).
- [`CLAUDE.md`](CLAUDE.md) orientation doc for future Claude Code instances: project identity, guardrails as one-liners, tier ordering, parent-project pointers, commands, dependencies & supply chain, open questions. Points at the design doc and ADRs rather than restating their content.
- Claude Code configuration under `.claude/`: project-level
allow/deny permissions and two skills — `skill-create` (generic
skill authoring guidance) and `git-commit` (this repo's commit
conventions: no Conventional Commits, Keep a Changelog discipline,
- **`lib/` + `bin/clide.dart`** — Dart core, shared by the app and by one AOT-compiled binary that has two modes: `clide <subcommand>` (one-shot for Claude) and `clide --daemon` (long-running for the app). Owns IPC, PTYs (via `ptyc`), subprocesses, file watchers, git shell-outs, pql invocations.
- **`lib/`** — all Dart code. Subsystem handlers (`lib/src/daemon/`, `lib/src/pty/`, `lib/src/ipc/`, `lib/src/git/`, `lib/src/pql/`), kernel services (`lib/kernel/`), UI widgets (`lib/widgets/`), built-in extensions (`lib/builtin/`), and the extension framework (`lib/extension/`). The Flutter app hosts the IPC server in-process (D-56).
- **[`pql`](https://github.com/postmeridiem/pql)** — external supporter tool. Clide wraps it for every query surface; never re-implements it.
- **`ptyc`** — small C supporter tool, peer of pql. Spawns a PTY + child and hands the master fd back over `SCM_RIGHTS`. Clide shells out to it for every pane (shell, tmux, claude, LSP, debug adapter).
- **`ptyc/`** — small C supporter tool, peer of pql. Spawns a PTY + child and hands the master fd back over `SCM_RIGHTS`. Clide shells out to it for every pane (shell, tmux, claude, LSP, debug adapter).
App ↔ daemon ↔ CLI speak a single JSON-lines unix socket protocol. The daemon outlives app restarts so Claude sessions survive reopens. Native rendering — markdown, canvas, graph — is Dart/Flutter (`CustomPaint` + widgets), not third-party packages.
tmux owns Claude session persistence (D-41) — the app re-attaches on restart via `tmux new-session -A`. Native rendering — markdown, canvas, graph — is Dart/Flutter (`CustomPaint` + widgets), not third-party packages.
Design doc: [`docs/initial-plan.md`](docs/initial-plan.md). Decisions: [`decisions/`](decisions/) (`D-NNN` confirmed, `Q-NNN` open, `R-NNN` rejected — see [`decisions/README.md`](decisions/README.md)). Python Textual predecessor under [`legacy/`](legacy/).
These are load-bearing. Violating any means the design is wrong, not the rule.
- **Flutter desktop is the host. No Electron, ever.** Web target may work as a happy accident — don't compromise desktop fidelity for it. If we ship a web build at all, prefer Flutter's **WebAssembly (CanvasKit/Skwasm) compile** over the JS/HTML renderer: it matches the desktop rendering pipeline, keeps our custom `CustomPaint` components pixel-identical, and avoids the DOM-renderer quirks around input handling and terminal-style content. `xterm.dart` is the terminal renderer (Tier 1); markdown, canvas, graph are custom `CustomPaint`/widget components (Tiers 2+, 5).
- **No heavy lifting in the UI layer.** The app renders and handles input; process/PTY/IO lifecycles live in the daemon. The daemon is Dart too — the split is *process boundary*, not language boundary.
- **CLI-first, not MCP.** Claude talks via Bash (`clide ...`), matching pql's contract. See [`D-001`](decisions/architecture.md#d-001-cli-first-not-mcp).
- **Dart is the core; native supporter tools fill specific gaps.** One Dart AOT binary for CLI + daemon.`ptyc` (C) for PTY spawning. `pql` (Go) for queries. No second "core language." See [`D-005`](decisions/architecture.md#d-005-dart-core-sidecar-dissolved-ptyc-as-pql-peer) (supersedes [`R-002`](decisions/rejected.md#r-002-go-sidecar)).
- **Own the rendering stack.** PTY (via `ptyc`), markdown renderer, graph, canvas — all clide-owned, not pulled from opinionated packages. Third-party rendering is where we'd hit ceilings first; we'd rather pay the cost up front.
- **User/Claude parity.** Every CLI subcommand has a UI affordance in the app, and every UI action has a CLI. Events are symmetric: every UI state change is a subscribable event. See [`D-006`](decisions/architecture.md#d-006-cli-and-event-surface-contract).
- **pql: wrap, don't duplicate, and treat it as a clide subsystem when present.** Pql logic only lives in `lib/src/pql/` (pure shell-outs). Clide owns pql's `ignore_files:` config key; it never touches pql's `.pql/` index/cache data. See [`D-003`](decisions/architecture.md#d-003-pql-as-supporter-tool-clide-wraps-never-duplicates).
- **Repo-is-the-workspace.** The git repo root is the workspace — no parallel "vault" concept. Clide dogfoods against its own repo.
- **Ignore discipline.** Single knob: `ignore_files:` in `.pql/config.yaml`, ordered layering. Default `[.gitignore]`; clide writes `[.gitignore, .clideignore]` when `.clideignore` exists. See [`D-004`](decisions/architecture.md#d-004-ignore-file-strategy).
- **Decision discipline.** All architectural choices live in `decisions/<domain>.md` as `D-NNN` records. Open questions live in `decisions/questions-<domain>.md` as `Q-NNN`. Rejected alternatives live in `decisions/rejected.md` as `R-NNN`. Before an architectural change, read the relevant domain file; before disagreeing with a guardrail, propose an amendment to the underlying `D-NNN` rather than a one-off. Claim new IDs via `tools/scripts/plan decisions claim D <domain> "title"` (until pql ships planning subcommands — see [`Q-021`](decisions/questions-architecture.md#q-021-pql-absorbs-planning-vs-keeps-separate)). See [`decisions/README.md`](decisions/README.md).
- **Flutter desktop is the host. No Electron, ever.** Web target may work as a happy accident — don't compromise desktop fidelity for it. If we ship a web build at all, prefer Flutter's **WebAssembly (CanvasKit/Skwasm) compile** over the JS/HTML renderer. `xterm.dart` is the terminal renderer; markdown, canvas, graph are custom `CustomPaint`/widget components.
- **Single process.** The Flutter app hosts everything in-process: IPC server, subsystem handlers (pane, files, editor, git, pql), extensions. No separate daemon binary (D-56 dissolved it). The CLI surface for Claude is a thin C client (ptyc peer).
- **CLI-first, not MCP.** Claude talks via Bash (`clide ...`), matching pql's contract. See [`D-1`](decisions/architecture.md#d-1-cli-first-not-mcp).
- **Dart is the core; native supporter tools fill specific gaps.** `ptyc` (C) for PTY spawning. `pql` (Go) for queries. No second "core language." See [`D-5`](decisions/architecture.md#d-5-dart-core-sidecar-dissolved-ptyc-as-pql-peer) (amended by D-56).
- **Own the rendering stack.** PTY (via `ptyc`), markdown renderer, graph, canvas — all clide-owned, not pulled from opinionated packages.
- **User/Claude parity.** Every CLI subcommand has a UI affordance, and every UI action has a CLI. See [`D-6`](decisions/architecture.md#d-6-cli-and-event-surface-contract).
- **pql: wrap, don't duplicate.** Pql logic only lives in `lib/src/pql/` (pure shell-outs). Clide owns pql's `ignore_files:` config key; it never touches pql's `.pql/` index/cache data. See [`D-3`](decisions/architecture.md#d-3-pql-as-supporter-tool-clide-wraps-never-duplicates).
- **Repo-is-the-workspace.** The git repo root is the workspace — no parallel "vault" concept.
- **Ignore discipline.** Single knob: `ignore_files:` in `.pql/config.yaml`, ordered layering. See [`D-4`](decisions/architecture.md#d-4-ignore-file-strategy).
- **Decision discipline.** All architectural choices live in `decisions/<domain>.md` as `D-NNN` records. Open questions as `Q-NNN`. Rejected alternatives as `R-NNN`. Claim new IDs via `pql decisions claim D <domain> "title"`. See [`decisions/README.md`](decisions/README.md).
See `docs/initial-plan.md` for the full tier definitions and acceptance criteria.
## Parent projects
- **`legacy/`** — Python Textual clide v1.2.0. Feature-frozen. Reference for the pane model, panel set, git skills (`/commit`, `/stash`, `/pull`, `/push` — rewire to `clide git …`), TODO.md parsing format.
- **`projects/claudian`** (April 2026, discarded) — 2-day experiment with an Obsidian-plugin approach. Its architectural patterns (CLI-first, pql-as-subsystem, ignore-file strategy, supply-chain gate, changelog discipline, commit conventions) are the records you see under `decisions/` and the skills under `.claude/skills/`. It first proposed a Go sidecar; [`D-005`](decisions/architecture.md#d-005-dart-core-sidecar-dissolved-ptyc-as-pql-peer) reversed that in favour of a Dart core.
- **[`projects/pql`](https://github.com/postmeridiem/pql)** — active supporter tool. Clide depends on it; never duplicates it.
- **Dart (core + app):** prefer-zero-deps. Flutter-SDK widgets first; third-party packages need justification. What stays is exact-pinned in `pubspec.yaml` (no caret ranges). Advisories reviewed before every bump; `pubspec.lock` committed.
- **Prefer-zero-deps.** Flutter-SDK widgets first; third-party packages need justification. What stays is exact-pinned in `pubspec.yaml` (no caret ranges). Advisories reviewed before every bump; `pubspec.lock` committed.
- **Document every bundled dependency.** Listed in [`assets/licenses.yaml`](assets/licenses.yaml) with name, kind, version, homepage, license, and purpose. Adding a dep is a two-step commit: add the artefact **and** the `licenses.yaml` entry. See [`D-42`](decisions/tooling.md#d-42-bundled-dependencies-documented-in-licensesyaml).
- **`ptyc` and any future native supporter tool:** no dep graph by design (libc-only for `ptyc`). "Audit" is reading the source before each bump.
-`pubspec.lock` is always committed.
-`make security` runs the Dart advisory review; `ci/security.sh` is the CI entry.
## Commands
(Placeholder until Tier 0 commits land these targets in real form.)
```
make build # dart compile exe bin/clide.dart -o bin/clide
make test # flutter test
make test-integration# daemon + CLI + fixture-repo suite
make run # launch Flutter desktop app
make analyze # flutter analyze
make format # dart format --set-exit-if-changed
make test # fast test suite (analyze + format + unit + widget + golden)
make test-integration# real app boot integration tests
make build-linux # flutter build linux
make build-macos # flutter build macos
make ptyc-build # build the ptyc PTY-spawn helper
make security # Dart advisory review + ptyc source review
make push-check # pre-push gate: analyze + format + test
make push-check # pre-push gate: decisions + core + fast tests + a11y
make hooks # install the repo's git hooks (one-time setup)
make clean # remove build artefacts
```
@@ -78,7 +80,7 @@ One-time setup on a fresh clone: `make hooks && flutter pub get` once Flutter is
## Changelog discipline
[Keep a Changelog 1.1.0](https://keepachangelog.com/en/1.1.0/). Every user-visible commit adds an entry under `## [Unreleased]` in [`CHANGELOG.md`](CHANGELOG.md). Cutting a release means moving Unreleased entries under a new dated version heading **and** bumping `project.yaml``version:` in the same commit — see [`.claude/skills/git-commit/SKILL.md`](.claude/skills/git-commit/SKILL.md) for the full rule.
[Keep a Changelog 1.1.0](https://keepachangelog.com/en/1.1.0/). Every user-visible commit adds an entry under `## [Unreleased]` in [`CHANGELOG.md`](CHANGELOG.md). Cutting a release means moving Unreleased entries under a new dated version heading **and** bumping `pubspec.yaml``version:` in the same commit — see [`.claude/skills/git-commit/SKILL.md`](.claude/skills/git-commit/SKILL.md) for the full rule.
build:## Build the Dart AOT `clide` binary (CLI + --daemon modes in one binary).
ifeq($(APP_PRESENT),yes)
dartcompileexebin/clide.dart-obin/clide\
--define=clideVersion=$(VERSION)\
--define=clideCommit=$(COMMIT)\
--define=clideDate=$(DATE)
.PHONY:run
run:## Launch the Flutter desktop app.
ifeq($(FLUTTER_OS),linux)
GDK_BACKEND=x11 LD_LIBRARY_PATH=$(CURDIR)/native/linux-x64$${LD_LIBRARY_PATH:+:$$LD_LIBRARY_PATH} flutter run -d linux --dart-define=CLIDE_PROJECT=$(CURDIR)
else
@echo"(pubspec.yaml not scaffolded yet; skipping)"
This document governs what clide is allowed to do at runtime, what it's allowed to depend on, and how contributors — human and agent — introduce code into the project. It is binding on all contributors. When in doubt, stop and ask.
Rationale for specific architectural choices referenced here and in code comments (the D-### markers) lives in `decisions/`. This document sets the rules; `decisions/` records why the rules produced the code they did in a given case. If the two ever disagree, the rule in this document wins until the document itself is changed.
## Why this document exists
clide's security model rests on a simple claim: the behavior of the app on a user's machine is fully determined by the signed release artifact and the files in this repository at the time of that release. Every file the app reads, every binary it executes, and every wasm component it instantiates on the default launch path was present, reviewed, and pinned when we built the release.
The moment that claim stops being true — the moment the app fetches something from the network that wasn't audited at build time — the entire sandboxing and trust story collapses. It doesn't matter how good the capability model around wasm components is if the components themselves are downloaded from the internet at first launch.
This policy exists to keep that claim true.
## The core rule: no network on the default launch path
clide does not perform network I/O during app startup, library initialization, or first use of any API, unless the user has explicitly asked for it.
"Explicitly asked for it" means the user pressed a button, ran a command, or otherwise took an action whose stated purpose is to cause a network fetch. Opening the app is not such an action. Opening a file is not such an action. Typing in a buffer is not such an action.
### What this prohibits
- Libraries that download their own native binaries on first import or first call (the `wasm_run` pattern).
- Language servers, formatters, or grammars that auto-install from the network when a file of that language is opened.
- Fonts, icons, or assets fetched from a CDN rather than bundled.
- Telemetry, crash reporting, or analytics that connect on startup regardless of user consent.
- Auto-update checks that run without the user asking.
- Any transitive dependency behavior that does the above, regardless of whether our code triggered it.
### What this allows, grudgingly
Signed, pinned fetches are permitted for a narrow set of cases where the alternative is materially worse. To qualify, all of the following must be true:
1. The artifact URL is hardcoded in the repository, not constructed at runtime.
2. The artifact is verified against a cryptographic hash committed to the repository, or a signature verified against a public key committed to the repository.
3. The fetch is cached; subsequent launches do not re-fetch.
4. Fetch failure produces a clear error, not a silent degradation.
5. The first launch after install can complete its primary function without the fetch succeeding. If the fetch is required for the app to work at all, this is not a "grudging allowance" case — vendor it instead.
If you cannot satisfy all five, the answer is vendoring or explicit user action, not a signed fetch.
### What this allows freely
Network I/O is unrestricted when the user has explicitly initiated it. Examples:
- "Install the Rust grammar" — fetches on button press, fine.
- Cloning a repository the user typed the URL of.
- Connecting to an LSP server the user configured.
- Any feature whose stated purpose is to talk to the network (HTTP client, git, SSH).
The rule is about the *default* path, not about forbidding network ever.
## Dependencies: zero by design
clide's default answer to "should we add a dependency" is **no**. Every dependency is a liability — code we didn't write, running in our users' address space, that we are nonetheless responsible for. The right number of dependencies is the number that makes clide meaningfully better than writing the code ourselves, and that number is small.
This is not a preference. It is the project's architectural stance, documented in `CLAUDE.md` as "no dep graph by design, exact-pinned, CVE-checked, audited." Every dependency that exists in clide is there because someone argued for it, someone reviewed the argument, and the answer was yes. Dependencies do not accumulate by accident.
### What "no dep graph by design" means in practice
- **Prefer writing it.** A 100-line utility written in-repo is cheaper than a 100-line dependency, even though the dependency looks free. We own the code either way; the dependency just hides that fact behind a pub.dev page.
- **Prefer the standard library.** Dart, Flutter, and Rust all have large standard libraries. Reach for them before reaching for a package. "There's a package that does this" is not a reason; "the standard library cannot do this reasonably" is.
- **Prefer inlining over depending.** If a package is small and does one thing we need, copy the relevant code (with attribution) rather than taking on the dependency. License permitting, this is usually the right call for utilities under a few hundred lines.
- **Prefer vendoring over fetching.** When a dependency is unavoidable and small enough to own, fork it into the repo and maintain it ourselves. The dep graph stops at our fence.
- **Reject deep trees.** A direct dependency that itself has fifteen transitive dependencies is fifteen dependencies we are taking on, not one. Evaluate the whole subtree, not just the top.
- **Format engines clear the bar.** Packages that parse or render external file formats (SVG, markdown, HTML, terminal escapes, tree-sitter grammars) are adoptable — they are not shortcuts for lazy coding but maintained renderers for specs we didn't write. The distinction: UI chrome (panels, tabs, canvas, layout) we own; someone else's file format we adopt and sandbox. See D-58.
### Every dependency is audited
For every dependency that does land, the following must be true and documented:
- **Exact-pinned.** `1.2.3`, never `^1.2.3` or `>=1.2.0 <2.0.0`. Lockfiles are committed. Updates are deliberate, reviewed PRs, not automatic.
- **CVE-checked.** The dependency is checked against known vulnerability databases at the time of adoption and on every subsequent CI run via the pipeline's automated CVE gate. A dependency with unpatched high-severity CVEs does not land; a dependency that develops an unpatched high-severity CVE after it lands becomes a release blocker until resolved. The CVE check is part of the build, not a periodic chore — if CI is green, the dep tree was clean at build time.
- **Source-reviewed.** Someone on the project has read the source of the dependency, or of the relevant subset, before it lands. "I trust this maintainer" is not a substitute. For large deps, the review can be scoped to the parts we actually use.
- **Recorded.** The reason for adoption, the reviewer, the source SHA or version, and the alternatives considered are captured in the PR that adds it.
- **Justified in place.** Every dependency in `pubspec.yaml` (and equivalent manifests) carries an inline comment explaining what it does, why we took it rather than writing the code, and its license. PR descriptions are not a substitute — the justification must survive in the manifest where future contributors will actually look. If a dependency can't be justified in two or three lines next to its version pin, it probably shouldn't land.
### The "is this worth it" test
Before adding any dependency, answer honestly:
1. What specifically does this do that we cannot reasonably do ourselves?
2. How much code would we write if we did it ourselves? (If the answer is under ~200 lines, strongly prefer writing it.)
3. What is the full transitive dep tree we are accepting?
4. Who maintains it, and what happens to clide if they stop?
5. Is the dependency's scope larger than our need? (Most packages are — we take on all the code, not just the parts we use.)
A dependency that passes this test and the vetting checklist below is welcome. A dependency that passes neither is a no. A dependency that passes one but not the other is a conversation, not a commit.
## Dependency vetting checklist
Before adding any new dependency (direct or transitive) to clide, verify:
**Network behavior**
- Does the package perform network I/O during import, initialization, or first call? If yes, reject or vendor.
- Does the package have a `postinstall`, `setup`, or equivalent script that downloads binaries? If yes, reject or vendor.
- Does any transitive dependency do either of the above? `flutter pub deps` and read the code of anything unfamiliar.
**Binary provenance**
- If the package ships native binaries, are they built from source in the same repo, or fetched from a release artifact? If fetched, treat as a network-at-launch violation.
- If the package expects to download binaries on a different machine (CI, developer setup), that is also disqualifying for runtime deps.
**Maintainership**
- Single-maintainer packages require explicit sign-off and a documented fallback plan. Note the maintainer and last meaningful commit in the PR.
- Packages with no activity in 12+ months require either a fork we control or a decision to own the code ourselves.
**Surface area**
- Prefer packages that do one thing. A dependency that adds 15 transitive deps to solve a 100-line problem should be inlined instead.
**Version pinning, lockfiles, and CVE review**
- See "Dependencies: zero by design" above. All the requirements there apply: exact-pinned, CVE-checked, source-reviewed, justified in place, recorded in the PR.
-`pubspec.lock` (and equivalent lockfiles for other toolchains) is committed to the repository and regenerated on every dependency change. A stale lockfile is a supply-chain hazard: a dependency removed from `pubspec.yaml` that persists in `pubspec.lock` will continue to install in CI and on fresh clones. Lockfile regeneration is part of the dep-change PR, not a follow-up.
**License**
- See License and attribution rules below. A dependency with an incompatible or unclear license is disqualifying regardless of technical merit.
If you can't answer any of these questions, you haven't vetted the dependency yet.
## Removing a dependency
Removal is not the reverse of addition. Adding a dependency involves a single decision and a single commit; removing one involves finding every place the dependency or its transitive siblings have leaked into the codebase and making sure none of them outlive the removal PR.
When removing a dependency:
1.**Grep the entire repository** for references to the package, its exports, and any type names it contributed. `rg '<package>|<PackageType>|<prefix_>'` across the repo. Zero hits outside git history is the goal. A single lingering import will break the build; a single lingering FFI stub or type alias will compile fine and fail at runtime.
2.**Regenerate the lockfile** as part of the same PR. A `pubspec.yaml` with the dep removed but a `pubspec.lock` that still pins it is a partial removal, and CI or a fresh clone will happily continue installing the package.
3.**Update `app/assets/licenses.yaml`** to drop the removed package and any transitive deps it brought in that aren't pulled by anything else. If the license manifest is auto-generated on release, verify the generation script sees the change; if it's maintained by hand, edit it in the same PR.
4.**Remove any vendored artifacts** tied to the dep — binaries, prebuilt assets, generated bindings — and delete their `BUILD.md` records. An orphaned vendored binary is worse than a removed one because it looks legitimate.
5.**Check for architectural assumptions** that the dep was carrying. If the removed package was the thing that justified a specific data flow, build step, or platform strategy, either the replacement picks up those responsibilities or the architecture has actually changed and the relevant design decision (see `decisions/`) needs updating.
A dependency is not removed until all five are true. "I deleted the line from pubspec.yaml" is the start of the removal, not the end.
## Vendored binary rebuild process
clide ships pre-built native libraries (currently `libtree-sitter.so` and per-platform equivalents) vendored into the repository. These binaries are part of the trust boundary: the signed release contains exactly these bytes, and users running clide are running exactly this code.
Every vendored binary has a corresponding `BUILD.md` (or equivalent) next to it in the repo that records:
1. The exact upstream source (git URL and commit SHA, not a version tag).
2. The full build command, including all compile flags.
3. The toolchain version used (compiler, linker, target triple).
4. The expected output size and a SHA-256 hash of the resulting binary.
5. Any patches applied to the source, stored as `.patch` files in the same directory.
Rebuilds are deliberate events, not background maintenance:
- Rebuilds happen in CI, not on a contributor's laptop. A matrix job cross-compiles for all supported platforms and produces the full set of binaries from a single source SHA.
- The rebuild PR updates `BUILD.md`, the binaries, and the hashes in one atomic change. The hash change is the signal in review that binary content changed and must be re-reviewed.
- No binary is committed without a corresponding source SHA and build log. If you can't reproduce the binary from the recorded inputs, the binary is not trustworthy and must be rebuilt.
- Security patches to vendored dependencies are tracked with the same urgency as source-level vulnerabilities. "We're on an old libtree-sitter because rebuilding is annoying" is not acceptable.
Supported platforms and their binary artifacts are documented in the repository root. Dropping a platform requires a policy decision, not a build convenience. Adding a platform requires adding it to the CI matrix and rebuilding all vendored binaries for it before the platform ships.
## Telemetry and phone-home rules
clide does not phone home. The app does not report to any server — ours or anyone else's. This is not a "default off" setting; it is an architectural commitment. clide is a space to think, not a surface for data collection.
This means:
- No analytics SDKs. Not Firebase, not Sentry, not Mixpanel, not a hand-rolled metrics endpoint. Not in debug builds, not in release builds, not behind a feature flag.
- No crash reporters that upload automatically. Crashes produce local logs the user can read and, if they choose, attach to a bug report they submit manually.
- No "check for updates" that runs without the user asking. Update checks are a button.
- No license validation that calls home. clide is MIT; there is nothing to validate.
- No feature flags fetched from a server. Configuration is local.
- No A/B testing, no experiments, no remote config, no "anonymous usage statistics."
If a future contributor proposes telemetry under any framing — opt-in, anonymized, debug-only, "just errors," "just for us," "just to improve the product" — the answer is no. The architectural commitment is the feature. Users installing clide are choosing a tool that does not watch them, and that promise is worth more than any data we could collect.
This rule does not bend. Proposals to add telemetry are out of scope for this project, full stop. If you want telemetry in your editor, clide is not the editor for you, and that's fine — fork it.
## License and attribution rules
clide itself is MIT-licensed. Every dependency, vendored binary, bundled font, and bundled asset must be:
1. Compatible with clide's license.
2. Attributed in the about screen's license manifest.
3. Attributed correctly — the license text and copyright notice must match upstream, not a paraphrase.
Compatible with care (copyleft): MPL-2.0 is acceptable for libraries; LGPL is acceptable only for dynamically-linked vendored binaries where we can demonstrate users can replace the library.
Not compatible: GPL (any version) for linked code, AGPL for anything, SSPL, "commercial use prohibited" licenses, custom licenses not reviewed by someone qualified to review them.
When in doubt about a license, the dependency does not land until the question is resolved.
### Attribution requirements
- The license manifest at `app/assets/licenses.yaml` lists every dependency with its license, copyright notice, and upstream URL.
- Transitive dependencies are listed, not just direct ones. If `wasm_run` pulls in `wasmtime` which pulls in `cranelift`, all three appear.
- Apache-2.0 dependencies get their `NOTICE` file content preserved verbatim, not summarized.
- Apache-2.0-with-LLVM-exception (e.g., Cranelift, parts of LLVM) requires the LLVM exception text specifically, not just the Apache-2.0 boilerplate.
- Fonts and icon sets get attributed even if the license doesn't strictly require it. It's the right thing to do.
-`app/assets/licenses.yaml` is regenerated as part of the release build, not maintained by hand. A release that ships a stale manifest is a release defect.
Adding a dependency means updating the license manifest in the same PR. No exceptions.
## Changelog and commit conventions
clide follows [Keep a Changelog 1.1](https://keepachangelog.com/en/1.1.0/) for `CHANGELOG.md` and [Conventional Commits 1.0](https://www.conventionalcommits.org/en/v1.0.0/) for commit messages. Enforcement is handled by the project's git skill; this section exists so human contributors know the standard before their first PR, and so the connection between these conventions and the rest of the policy is explicit.
Security-relevant changes — CVE responses, dependency-driven vulnerability fixes, the removal of a phoning-home transitive dep, anything where the rules in this document were the reason for the change — go under the `Security` heading of the release's changelog entry, regardless of whether the code change itself looks security-shaped. That heading is the trail future-us follows to reconstruct why a dep was bumped or removed. Lumping security fixes under `Fixed` because the diff looks like a normal bug fix loses that signal and is the wrong choice even when it's technically accurate.
## When this policy is wrong
This document is a set of rules, but rules encode judgment calls that may not survive contact with a specific situation. If you think the policy is wrong for a specific case, the process is:
1. Stop. Do not ship the thing.
2. Write down the specific case and why you think the rule fails.
3. Propose a change to this document.
4. Get the change reviewed on its merits.
5. If the change lands, proceed. If not, the rule holds.
"The rule is inconvenient for this task" is not an argument for the rule being wrong. "The rule produces a worse outcome than the alternative, measured against the reasons the rule exists" is.
## Notes for agents
If you are an AI agent working on this repository, the following guidance is binding:
- The default answer to "should I add a dependency" is no. Write the code instead, or use the standard library. If you think a dependency is genuinely warranted, stop and make the case to the human before adding it — do not add it and justify it afterward.
- Do not add dependencies without walking the vetting checklist. Record your answers in the PR description.
- Do not add code that performs network I/O on the default launch path. If you're not sure whether a library does this, read its source before adding it, not after.
- Do not vendor binaries without creating the corresponding `BUILD.md` and recording the source SHA and build command. A committed binary with no reproducibility record is worse than no binary at all.
- Do not add telemetry, analytics, crash reporting, remote config, feature flags, or any other phone-home mechanism under any framing. This is not a "default off" setting; the project does not have telemetry and will not have telemetry. If you think you need it, you are wrong.
- When in doubt, stop and ask the human. Pivoting architectures in a single session to route around a principle is almost always wrong; the principle is there for a reason that may not be visible from inside the task.
A Flutter desktop IDE for Claude Code. Markdown-first content, pql-powered queries, canvas and graph surfaces, a Go sidecar handling PTYs / subprocesses / git / pql. Claude drives the UI through a `clide` CLI.
A Flutter desktop IDE for Claude Code. Native rendering, terminal-first interaction, pql-powered queries, canvas and graph surfaces. Linux and macOS.
Currently pre-v2.0, scaffolding. The north-star design lives in [`docs/initial-plan.md`](docs/initial-plan.md); architectural decisions are captured in [`docs/ADRs/`](docs/ADRs/). The Python Textual v1.2.0 implementation is archived under [`legacy/`](legacy/) for reference.
## Architecture
## Three surfaces, one tool
Single Flutter package at the repo root. The app hosts everything in-process: IPC server, subsystem handlers (pane, files, editor, git, pql), and the extension framework. tmux owns Claude session persistence.
- **`app/`** — Flutter desktop application (Linux / macOS; Windows is a stretch).
- **`sidecar/`** — Go sidecar/CLI, single binary with two modes: `clide <subcommand>` (one-shot for Claude) and `clide --daemon` (long-running sidecar for the app). Owns PTYs, subprocesses, file watchers, git, pql invocations.
- **[`pql`](https://github.com/postmeridiem/pql)** — external supporter tool. clide wraps it for every query surface; never re-implements it.
- **`lib/`** — all Dart code. Kernel services (theme, i18n, settings, panels, commands, focus), UI widgets, built-in extensions, and the extension contract.
- **`ptyc/`** — small C helper. Spawns a PTY + child and hands the master fd back over `SCM_RIGHTS`. Every pane (shell, tmux, claude, LSP, debug adapter) goes through it.
- **[pql](https://github.com/postmeridiem/pql)** — external supporter tool. Clide wraps it for every query surface; never re-implements it.
## Why the rebuild?
Claude drives the UI through a `clide` CLI surface (Bash, not MCP). Every CLI subcommand has a UI affordance and every UI action has a CLI equivalent.
The Python Textual implementation (`legacy/`) proved the pane model but capped at terminal-only rendering. A short-lived experiment (`claudian`, April 2026) explored an Obsidian-plugin approach and was abandoned because Obsidian is Electron, and the user's history with terminal-in-Electron ruled that host out.
## Built-in extensions
Flutter desktop resolves both constraints: native Skia rendering avoids Electron's failure modes, `xterm.dart` gives us a solid terminal surface, and the pane model can be rebuilt natively. The Obsidian ideas worth keeping (canvas, graph) fold in; everything else (vault concept, inline query tables, plugin ecosystem inheritance) doesn't.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.