jpmschweitzerandClaude Opus 5.5
ddce4441a9
tooling/db/ (a misnamed directory: connectors, not database work),
trellis-batch.sh and synth_ui_sounds.py become `reach assets`:
audio {health,generate,batch,post {convert,normalize,trim,pipeline}},
image {health,generate}, trellis {health,generate,batch}, and synth-ui.
The four audio bash wrappers are retired, and tooling/db/ is gone.
Parity, from baselines taken before anything moved:
- the four UI-sound WAVs and the harmonic-synth WAVs (exponential and linear
decay) are byte-identical
- the ffmpeg pipeline's decoded PCM is identical. Its .ogg bytes are not,
even between two runs of the OLD code: Ogg picks a random stream serial,
so the encoded file was never the right thing to compare
- the network success paths can't be run in a gate (Stable Audio and Trellis
are kept off, Gemini costs money), so tooling/test_assets.py stands up a
fake Gradio and pins every payload: the audio submit, Trellis's six-call
session sequence with its 9-input image_to_3d, and the Gemini body. It
failed when one Trellis value was mutated (7.5 → 7.0)
Failure classification, in endpoints.py, is the point of the port. The
services are OFF by design (VRAM on tower-of-joy, D-17), and the topology doc
warns against "fixing" one by restarting it. So a refused connection says OFF
and asks for the service to be turned on rather than restarted; a 4xx/5xx says
the request was rejected; 401/403 says credentials; 429 says quota; and an
unreachable Gemini blames the network, not VRAM.
Behaviour changes, each a failure that used to read as success or crash:
- audio batch and trellis batch exited 0 with failures in their summaries;
they now print the summary and exit 1
- trellis generate on a missing image crashed with a TypeError
(print(..., indent=2)); it now names the file, and checks it before the
service so a typo is not reported as an outage
- the ffmpeg pipeline left its intermediates behind when a step failed
Structure: the connectors called each other as subprocesses (batch spawned
the connector, which spawned audio_post) and parsed each other's stdout. They
are now function calls, and ffmpeg is the only exec, through core/process.
ensure_venv() is removed: it os.execv'd into .venv, which D-263's exec rule
forbids, and reach declares the dependencies itself. config.json moved into
the domain deliberately, and the local-services rule follows it.
Output contract: results are still JSON on stdout with the same keys, so skill
readers keep working. Failures are an exit status with a Fix line, never
{"ok": false}. The audio-gen, glb-gen and image-gen skills, Araminta's agent
file and the allow-list are updated to match. glb-gen's "trellis-batch.sh is
hardcoded to one category" caveat is gone: batch takes --input-dir or --names.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Settled Reach - Agent Team
Overview
The Settled Reach uses Claude Code agent teams to simulate a game development team. Each agent has a distinct personality, expertise area, and role. The team leader (Jeroen) directs the team through chat.
All agents read their briefing file at docs/briefings/{name}.md before starting work. Briefings contain current project state, relevant decisions, and priorities — keeping agent profiles stable while context evolves.
Agent Roster
Core team (brainstorming + design discussions)
| Agent | File | Role | Model |
|---|---|---|---|
miri |
miri.md | Worldbuilder & Setting Designer | sonnet |
ozzie |
ozzie.md | Player Experience / Wow Factor | sonnet |
paula |
paula.md | Narrative & Political Depth | sonnet |
gore |
gore.md | Themes & Endgame Design | sonnet |
gestalt |
gestalt.md | Systems Design & Fun Factor | sonnet |
nigel |
nigel.md | Sandbox & Replayability | sonnet |
tyre |
tyre.md | Technical Architecture | opus |
qatux |
qatux.md | Documenter & Librarian | sonnet |
Specialist team (task-focused, not in regular discussions)
| Agent | File | Role | Model | When to use |
|---|---|---|---|---|
troblum |
troblum.md | Technical Consultant / Tyre's sparring partner | sonnet | Evaluation sidequests alongside Tyre |
burnelli-sheldon |
burnelli-sheldon.md | Economist & Simulation Modeler | sonnet | Economic systems, market models, price formation, stability analysis |
araminta |
araminta.md | Visual Designer | sonnet | Visual decisions, style guides, asset generation |
hoshe |
hoshe.md | QA Engineer / Tester | sonnet | Testing, test plans, bug reports, verification |
Infrastructure team
| Agent | File | Role | Model | When to use |
|---|---|---|---|---|
si |
si.md | Refinement Manager | sonnet | Ticket context review before batch activation (/whats-next step 2) — active now |
clerk |
clerk.md | Institutional Guardrail | sonnet | Manual D-record/ticket consistency audits — dormant on the pre-push path (disabled by default, #965; see clerk.md) |
Implementation team (active)
| Agent | File | Role | Model | When to use |
|---|---|---|---|---|
dudley |
dudley.md | Server Developer | sonnet | Rust/bevy_ecs simulation: tick pipeline, ECS, world generation cascade, save/load — server/'s most active workstream |
stig |
stig.md | UI Developer | sonnet | Godot client UI: character customisation, insert/HUD, dialogue panels, fog/entity rendering |
justine |
justine.md | Polish & Deploy | sonnet | Build pipelines, packaging, performance, release prep, content-validation/pre-commit tooling |
mellanie |
mellanie.md | Copywriter | sonnet | In-game text, UI copy, tooltips, flavor text, player monologue |
Standby team (activate when implementation starts)
| Agent | File | Role | Model | When to use |
|---|---|---|---|---|
oscar |
oscar.md | Networking Developer | sonnet | Multiplayer networking, sync, client-server protocol |
inigo |
inigo.md | Sound Designer | sonnet | Soundscape design, ambient layers, diegetic cues, audio propagation (D-018) — activate when sound design starts |
Whether oscar (multiplayer) should exist yet for a Phase-4, pre-game project is an open question — see the team lead. Localization was decided out of scope (template-writing + Gemma integration prohibitively expensive), so the former tiger persona was removed (see governance/rejected).
Usage Modes (Hybrid Approach)
1. Single session - for brainstorming & discussion
The main Claude session plays all core team agents in conversation, switching voices as appropriate. Agent files serve as personality references. This is the default mode for design discussions.
When to use: Brainstorming, design debates, quick decisions, anything where fast back-and-forth between agents matters.
2. Subagent delegation - for focused tasks
Spawn individual agents as subagents for specific work:
Use tyre to evaluate Godot vs Bevy for our requirements.
Use miri to verify the canon accuracy of our Guardian faction design.
Use qatux to update DECISIONS.md with today's discussion.
Use araminta to create the initial color palette and style guide.
Use hoshe to write a test plan for the LOS system.
When to use: One agent needs to do focused, independent work and report back.
3. Agent teams - for parallel work
Spawn multiple agents as teammates with shared task list:
Create a team: spawn tyre and troblum as teammates.
Tyre: evaluate Godot against our requirements.
Troblum: evaluate Bevy against our requirements.
Compare findings when both are done.
Create a team: spawn gestalt, paula, and miri as teammates.
Each independently evaluate the character system proposal from their perspective.
Synthesize findings.
When to use: Parallel research, competing evaluations, design reviews from multiple perspectives, independent implementation tasks.
Mode selection guide
| Task type | Mode | Why |
|---|---|---|
| Design brainstorming | Single session | Fast, cheap, good voice consistency |
| Quick lore check | Subagent (miri) | Focused, returns answer |
| Engine evaluation | Team (tyre + troblum) | Parallel research, sparring |
| Design review | Team (core agents) | Independent perspectives, genuine disagreement |
| Style guide creation | Subagent (araminta) | Focused creative work |
| Test writing | Subagent (hoshe) | Focused, spec-driven |
| Implementation batch | Team (tyre + hoshe + relevant others) | Parallel build + test |
| Documentation update | Subagent (qatux) | Structured, accurate citations |
Agent usage notes
Troblum (Technical Consultant)
- Solo for performance profiling, benchmarking, and technology stress-tests; paired with Tyre for architecture second opinions — never the sole voice on an architecture decision
- Spawned for specific evaluation sidequests, not open-ended discussion
- Blunt, data-driven, will challenge Tyre's assumptions with evidence
Araminta (Visual Designer)
- Joins discussions only when visual consistency decisions are needed
- Drives image-gen/sprite-gen/glb-gen/audio-gen via Bash — no Skill/MCP tool grant, so she invokes the connectors directly, e.g.
reach assets image generate ... - image-gen calls the paid Gemini API (
GEMINI_API_KEY) — always ask Team Leader for permission before generating. glb-gen/audio-gen run against self-hosted tower-of-joy infrastructure and sprite-gen renders locally, so they don't carry the same per-call cost, but confirm intent before large batch jobs.
SI (Refinement Manager)
- Spawned by
/whats-nextto review ticket context before batch activation - Reads D/Q-records, workshop outcomes, existing code to assess ticket readiness
- Reports READY (with context summary) or GAPS (with specific ambiguities)
- Does not implement — refines
Qatux (Documenter & Librarian)
- Core team member — mandatory participant in every team interaction and discussion round (she is the record-keeper; a round without her produces no durable documentation)
- Maintains governance/ domain files, docs/discussions/ round records, and briefings
- Answers "did we discuss this?" with citations
Extending the team
To add a new agent:
- Create a
.mdfile in this directory - Use YAML frontmatter with at minimum: name, description, tools, model. Add
memory: projectonly for project-memory/librarian roles that persist learnings across sessions (currently just qatux) — one-shot task agents omit it - Write a personality prompt in the markdown body
- Add
Read your briefing at docs/briefings/{name}.md before starting work.to the profile - Create a briefing file in
docs/briefings/{name}.md - Add to the appropriate team table above
- Update TEAM.md
Project documents
All agents read their briefing at docs/briefings/{name}.md for current context. Key project documents:
governance/- Domain-split decision files (see governance/README.md for index)docs/discussions/- Active + archived discussion rounds (round-NN-topic.md, indexed indocs/discussions/README.md)TEAM.md- Team roster and role descriptionsCLAUDE.md- Project-wide conventions