Supports design discussion tickets that run on the planning branch
with a purpose-assembled agent panel. Includes Qatux (documenter)
and SI (project manager) for bookkeeping alongside domain agents.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add 5 new rules to the agent spawn prompt: read-before-write,
verify-after-write, no-partial-work, message-when-blocked, and
backward-compatibility. Adds verification checklist before marking
tasks done. Addresses recurring issues with agents skipping call
site updates, leaving partial implementations, and not escalating
blockers.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pass PR description inline instead of writing to /tmp/pr-body.md
and cat-ing it back — matches the tea-comment wrapper approach.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move CLI reference material, endpoint configs, and orchestration
patterns to .claude/rules/ (auto-loaded). CLAUDE.md now focuses on
project identity, safety rules, and workflow conventions.
Update cross-references in ticket, docs-search, and git-commit
skills to point to new rules/ locations.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduces 5 rules files following Anthropic's recommended pattern
for splitting large CLAUDE.md files. Rules auto-load every session
and separate CLI reference material from core project identity.
- tea-cli.md: Gitea CLI reference, PR creation, protected branches
- git-safety.md: staging rules, commit conventions (from git-commit skill)
- project-structure.md: full annotated directory tree
- team-patterns.md: model selection, large content push pattern
- local-services.md: Gitea/Qdrant/Ollama endpoints
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Reviews must run from main. Added step 0 hard stop if current branch
is not main, preventing repeated misuse from team branches.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Single-command wrapper that handles temp file creation and cleanup
so tea comment works reliably with multi-line strings. Pre-approved
in settings.json, referenced in CLAUDE.md and pr-review skill.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Delegate sprint-status to a haiku subagent (sprint-plan pattern) so
intermediate artifacts (sweep JSON, template read, PR list) stay out
of the main context window. Trim sweep JSON output by removing unused
fields (ok, sprint.status, priority, ticket_id) and shortening issue
detail strings. Condense output template by moving rendering rules
into SKILL.md and simplifying the bookkeeping table to two columns.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Orchestrates sprint sweep + tea pr list to produce a consistent
health report: tickets by status, PR cross-reference, bookkeeping
issues, and open work by team summary.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sprint-start: add step 9 (post-work lifecycle) — commit, push, review,
fix-comments loop, approve, then shutdown. Team stays alive until PR is
accepted. Step 8e updated to reference the new lifecycle.
pr-push: add step 9 (team awareness) — don't shut down agents after
pushing, suggest /pr-review next.
pr-review: add step 8 (post-review team actions) — cross-references
sprint-start step 9c for dispatching review comments to agents on
CHANGES_REQUESTED and shutdown on APPROVED.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All reviewer agents now read source files from worktree paths using
the Read tool instead of git show. Removes Bash dependency for
non-code reviewers and ensures files are always the actual branch
checkout.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace individual API call workflow with JSON-as-truth architecture.
Local wireframe JSON files are the source of truth; Frame0 is treated
purely as a renderer. New frame0-sync.py handles push/pull/export with
stable local ID <-> ephemeral Frame0 ID mapping via .idmap.json files.
Key changes:
- frame0-sync.py: push/pull/export with topo-sorted shape creation
- frame0-export-batch.sh: batch export with staleness check
- frame0-cmd.sh: fixed command names, list-pages JSON bug, added commands
- Removed frame0-wireframe.sh (superseded by JSON workflow)
- SKILL.md: renderer-only guidance, simplified workflow
- api-reference.md: corrected commands, type/color token mapping
- component-library.md: all 6 patterns rewritten as JSON templates
- .gitignore: added *.idmap.json
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replaces Frame0 MCP server with portable bash+curl scripts that call
Frame0's local HTTP API directly. Includes low-level API wrapper,
high-level composition helpers with project styling, component library
patterns, and setup guide. Requires Frame0 desktop app running locally.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Skill for triaging in-game bug reports from the Godot client's
user://bug-reports/ directory. Reads snapshot, render, inputs,
and description files; presents summaries; offers investigate,
create ticket, or dismiss actions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Spawned agents are told not to run git commands (team lead handles those)
and to call db/connectors scripts with bare commands only — no python3
prefix or command chaining — so they match the permission whitelist.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add sprint-plan, pr-push, pr-review, ticket, docs-search, and
workshop-start to shared permission settings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Case A on main previously closed the active sprint immediately without
checking ticket completion or open PR status. Now checks both and asks
the user before proceeding.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Implements 4 completed tickets + partial progress on 2 more:
- #93 Tier marker components (ActiveSim, BackgroundSim, StateSaved + TierPlugin)
- #138 Information tag schema (ObserverAccess enum in knowledge/types.rs)
- #124 Sound event system (SoundEventEmitter, SoundEventQueue, bridge wiring)
- #193 Line previewer CLI (line_preview binary with filter/explain/sequence modes)
- #94 Active tier simulation (in progress — With<ActiveSim> filters)
- #139 Component-level access control (in progress — filter_by_access)
Updates snapshot fixtures and test golden files for new sound_events field.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adopts the whatsinagame naming pattern where the domain comes
first and the action second (e.g. pr-review, sprint-start,
audio-gen). Updated all cross-references in settings, agents,
docs, and inter-skill references.
12 renames: commit→git-commit, create-skill→skill-create,
gen-audio→audio-gen, gen-image→image-gen, plan-sprint→sprint-plan,
push-pr→pr-push, render-sprite→sprite-gen, review-pr→pr-review,
search-docs→docs-search, start-sprint→sprint-start,
start-workshop→workshop-start.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Gitignored .tmp/ avoids Bash permission prompts when writing
temp files for tea comment posting. Added sed -n blanket
permission. Updated review-pr skill to use Write tool into
.tmp/ instead of Bash heredocs to /tmp/.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
push-pr now runs /commit first when uncommitted changes are detected,
making it a one-stop commit→push→PR command. Added Skill to
allowed-tools for cross-skill invocation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Documents audio-batch manifest format, synth parameters, and the
--post flag as the preferred workflows. Batch reduces approval
count from ~30 to 2 for multi-asset generation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
On main, the skill now assesses sprint state and does the next right
thing: close active sprint (Case A), activate planned sprint (Case B),
or suggest /plan-sprint (Case C). Cases chain automatically — closing
flows into activation if a planned sprint is waiting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Technical project descriptor with game architecture, simulation
parameters, and content model. Version field (0.1.9) is now the
canonical version — Cargo.toml synced, CLAUDE.md and start-sprint
skill updated to reference it during release workflow.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Step 7 now transitions all unblocked sprint tickets to in_progress
immediately rather than deferring to agents (who forget). Completes
the ticket lifecycle: start-sprint → in_progress, push-pr → review,
sprint stop → done.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Step 7 scans commit messages for #NNN ticket refs and transitions
in_progress tickets to review status. Only in_progress → review;
other statuses are left alone.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When invoked on main, /start-sprint now closes the active sprint,
bumps the version to v0.1.N, moves changelog entries under the new
version heading, tags, and pushes. Team branch behavior unchanged.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Documents the {domain}-{action} naming pattern for new skills
so future skills follow the convention established in the kit.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Agents were sending outputs as messages instead of writing files,
which prevented Qatux (documenter) from reading other agents' work.
Now each round task instructs agents to write to
docs/workshops/{name}/{agent}-roundN.md, and between-round steps
verify files exist before proceeding.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Agents were navigating outside the git root into sibling worktrees.
Added explicit boundary rules: all work stays within the git root
unless directly instructed otherwise. Updated plan-sprint to enforce
worktree-relative paths in generated briefings.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move universally-needed permissions from per-worktree
settings.local.json into the shared .claude/settings.json:
git operations, ticket/sprint/sqlite CLI, make, tea, skills.
Add deny rules for destructive operations (force push, hard
reset, clean -f, rm -rf). Migrate from deprecated :* suffix
to modern space-wildcard syntax.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Unified sprint script at db/connectors/sprint with 5 subcommands:
status, start, stop, start-work, prepare. Calls ticket CLI via
subprocess (no SQL duplication), auto-detects sprint from DB state
and team from git branch. Guards prevent activating unplanned
sprints. Updated start-sprint, plan-sprint, and ticket skills to
reference the new CLI. Added sprint CLI section to CLAUDE.md.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The skill previously relied on the agent hardcoding branch names,
which missed non-standard branches like planning. Now explicitly
instructs the agent to use git worktree list for discovery.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
