Use when manage multiple local CLI agents via tmux sessions (start/stop/monitor/assign) with cron-friendly scheduling.
日本語の概要は準備中です。原文の説明を表示しています。
Orchestrates the tdmcp feature team end-to-end: design/wireframe → build → integrate → QA → release. Use whenever the user wants to build, implement, develop, ship, or add one or more tdmcp features/tools/effects/controls/AI-prompts for TouchDesigner, or to run them through a coordinated design→develop→QA→deploy pipeline. Also use for follow-ups: re-run, continue, update, fix, or improve a feature/batch, ship the next version, or rebuild just one part of a previous run. Single feature or a whole batch.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Coordinate the five tdmcp specialists to take feature ideas from spec to a pushed release, following the project's proven workflow: parallel isolated builders + a single-writer integrator + live-validated QA + an autonomous release.
| Stage | Mode | Why |
|---|---|---|
| Design | sub-agent (fan-out if many features) | the architect is an isolated one-shot producer; specs land in _workspace/ |
| Build | sub-agent (fan-out, parallel) | builders are fully isolated by design (new files only, no inter-comms) — the textbook sub-agent case |
| Integrate → QA → Release | agent team | producer↔reviewer feedback loops (QA ↔ builder-fixer/integrator) and the release gate need live messaging |
Why this split: builders must NOT touch shared files so they can run concurrently; that isolation removes any need for build-time team comms. The integrate/QA/release trio, by contrast, lives on a tight fix-and-re-verify loop, which is exactly what an agent team is for.
| Agent | Type | Skill | Output |
|---|---|---|---|
td-architect | custom | td-feature-design | _workspace/01_design_<feature>.md |
td-builder (×N) | custom | td-feature-build | new tool + test files; _workspace/02_build_<feature>.md |
td-integrator | custom | td-feature-integrate | wired index.ts/agent.ts; _workspace/03_integrate.md |
td-qa | custom | td-feature-qa | _workspace/04_qa_<batch>.md (PASS/FAIL/UNVERIFIED) |
td-releaser | custom | td-feature-release | CHANGELOG + version bump + commit/tag/push; _workspace/05_release.md |
All Agent / TeamCreate calls use model: "opus".
_workspace/ exists._workspace/ → fresh run. Go to Phase 1._workspace/ exists + user asks to fix/continue/re-run part → partial re-run. Re-invoke only the affected agent(s), passing the prior artifact paths so they refine rather than rewrite._workspace/ exists + a new batch of ideas → new run. Move the old _workspace/ to _workspace_<YYYYMMDD_HHMMSS>/, then Phase 1.docs/ROADMAP.md._workspace/ and write the feature list to _workspace/00_input/features.md.Spawn one td-architect per feature (or one for the whole small batch) via Agent, subagent_type: "td-architect", model: "opus", run_in_background: true for parallelism. Each writes _workspace/01_design_<feature>.md. Wait for all, then read the specs and resolve any flagged overlaps/contention before building.
Spawn N td-builder sub-agents in a single message (run_in_background: true), one per spec. Each prompt includes its spec path and the hard rule: new files only, never edit shared registries/CLI/docs. Each returns green-in-isolation files (vitest + biome) and a build note. Size N to the batch; keep it to a manageable parallel set (~3–5 at a time for a large batch, then a second wave).
Form the team and run the convergence loop:
TeamCreate(team_name: "tdmcp-delivery", members: [
{ name: "integrator", agent_type: "td-integrator", model: "opus" },
{ name: "qa", agent_type: "td-qa", model: "opus" },
{ name: "builder-fixer", agent_type: "td-builder", model: "opus" },
{ name: "releaser", agent_type: "td-releaser", model: "opus" }])
(builder-fixer closes QA's fix requests on handler/schema/test bugs without re-spawning.)TaskCreate the pipeline with dependencies:
integrator)qa, depends on integrate) — not one big pass at the endbuilder-fixer/integrator, depends on QA findings)releaser, depends on QA = PASS for everything shipping)SendMessage: QA sends file:line + fix to the owner the instant a defect is found; boundary bugs go to both sides; QA re-validates after each fix (cap ~2–3 rounds/feature).TaskGet, intervenes if a member stalls._workspace/05_release.md exists (or that release was intentionally held).TeamDelete. Preserve _workspace/ for audit.[leader] → architect(s) ─ specs → builders (parallel) ─ files+notes →
┌─ TeamCreate(integrator, qa, builder-fixer, releaser) ─┐
│ integrator wires → qa validates (live if bridge up) │
│ ↑ fix requests (SendMessage) ↓ │
│ builder-fixer / integrator patch → qa re-checks │
│ qa PASS → releaser cuts + pushes │
└───────────────────────────────────────────────────────┘
↓
_workspace/0*_*.md + pushed tag
| Situation | Strategy |
|---|---|
| One builder fails in isolation | Ship the rest; route its spec to builder-fixer in Phase 4 (or re-spawn). Don't block the batch. |
| Build breaks on integrate | Integrator bisects, wires the rest, sends the offender a precise fix; team continues. |
| QA finds a boundary bug | SendMessage to both producer and consumer; re-validate after fix; cap ~2–3 rounds, else report blocker. |
| Bridge offline | QA runs offline gates, marks live checks UNVERIFIED-pending; release may still ship (note it) — never fail the pipeline for a missing TD. |
| QA = FAIL on a feature | Releaser holds it for next cycle; ships only PASS features; report what was held. |
| Concurrent agent's WIP breaks compile project-wide | Validate the slice in isolation (vitest run <file>); wait for their tree to go green before the full build/release. |
Normal: user asks to build the cue-sequencer feature → Phase 1 lists 1 feature → architect writes the spec → 1 builder produces a green tool+test → team forms → integrator wires + builds green → QA validates live (preview + post-cook errors clean) = PASS → releaser bumps minor, writes CHANGELOG, commits/tags/pushes → report names the new version.
Error: in a 4-feature batch, builder-3's tool cooks to a TD error caught by QA (a Level TOP gain no-op) → QA SendMessages builder-fixer with file:line + use brightness1 → fix re-validates PASS → the other 3 already PASS → releaser ships all 4. If builder-3 can't be fixed in 3 rounds, releaser ships the 3 PASS features and the report holds builder-3 for next cycle.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Use when manage multiple local CLI agents via tmux sessions (start/stop/monitor/assign) with cron-friendly scheduling.
日本語の概要は準備中です。原文の説明を表示しています。
Use when a hybrid memory system that provides persistent, searchable knowledge management for AI agents (Architecture, Patterns, Decisions).
日本語の概要は準備中です。原文の説明を表示しています。
Reference for how BDB structures autonomous software engineering work — the seven-node dispatcher graph (Architect, TechLead, UI/UX, Engineering, Media/EventTech, Reviewer, Shipping) that /startcycle-graph actually runs. Use when you need the high-level lifecycle framing without inventing your own process.
日本語の概要は準備中です。原文の説明を表示しています。
Tools are how AI agents interact with the world. A well-designed tool is the difference between an agent that works and one that hallucinates, fails silently, or costs 10x more tokens than necessary. This skill covers tool design from schema to error handling.
日本語の概要は準備中です。原文の説明を表示しています。
Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.
日本語の概要は準備中です。原文の説明を表示しています。
agenttrail: live map of a multi-agent build in the browser: which plan component is being worked on, by which agent or harness, what is done and what is stuck. Use when a multi-agent pipeline starts (/startcycle, /startcycle-graph, /teamwork-preview) or after a plan-canvas approve, when the user asks to see what the agents are doing ("live map", "build board", "who is running now"), or by yourself whenever a multi-agent run is under way — no user prompt needed.
日本語の概要は準備中です。原文の説明を表示しています。