メニュー
Select and run a multi-agent workflow pattern for work that exceeds one agent's reliable scope. Use for parallel research, independent reviews, cross-domain changes, or high-risk implementation planning.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use this skill whenever a decision is made about architecture, dependencies, frameworks, naming conventions, or layer order. Creates a numbered ADR (Architecture Decision Record) in `.harness/docs/adr/` in the canonical Nygard format. Always invoke this before changing layer order, adding a layer, swapping a major dependency, or introducing a new external service.
Use this skill whenever the user asks to add, implement, or build a new feature, capability, endpoint, page, command, or anything user-visible. Enforces the Anthropic two-fold harness pattern — read .harness/feature_list.json, pick exactly one feature, implement incrementally, run the structural test on every save, and never declare "done" without updating the JSON. Always invoke this skill instead of writing new feature code freehand.
Use this skill whenever the user reports a bug, unexpected output, or "this doesn't work". Runs the dev server, drives the failing flow via Playwright MCP if installed (else captures stdout/stderr), and produces a minimal repro before any fix. Mirrors the OpenAI Chrome-DevTools-Protocol-into-runtime pattern at solo scale — verify the failure before you propose a fix.
Use this skill whenever the user asks to analyze, audit, review, summarize, produce a report, write a plan, make a proposal, draft a decision doc, list "next actions", or any other task that produces a DOCUMENT a HUMAN reads-and-acts-on. Outputs a self-contained <slug>.html at repo root using the shared dark-theme CSS. Why HTML and not Markdown: golden principle #9 — Markdown is great for files an agent reads-and-edits (CLAUDE.md, SKILL.md, ADRs), but a HUMAN reading a 700-line MD deliverable will scroll, miss the conclusion, and ask the agent to clarify — burning turns and tokens. HTML deliverable is read once, decided once. Do NOT use this skill for files the agent itself reads (those stay MD), for stdout output from /review-this-pr or /garbage-collection (pass-through MD), or for short summaries under ~30 lines (overhead not worth it).
Use this skill weekly, before releases, or when the user mentions "stale docs", "doc drift", "docs are wrong", or "the README is out of date". Cross-checks every code path, file path, and command referenced in `.harness/docs/` and `CLAUDE.md` against the current repo state and produces a list of stale references — the doc-gardening agent pattern.
Use this skill whenever a skill, subagent, or hook is changed, before merging to main, or when the user mentions "eval", "regression test for the harness", or "is the harness still working". This skill is a thin wrapper — it runs a single shell command (`npm run harness:eval` or `python -m harness.eval_runner`) and summarizes the JSONL output. Do not implement the eval logic yourself; the runner is already deterministic.
| name | orchestrate |
| description | Select and run a multi-agent workflow pattern for work that exceeds one agent's reliable scope. Use for parallel research, independent reviews, cross-domain changes, or high-risk implementation planning. |
| allowed-tools | Read, Grep, Glob, LS, Bash(node .claude/skills/orchestrate/orchestrate.mjs:*), Bash(node .harness/scripts/orchestration-contract-from-task.mjs:*) |
| suggested-turns | 10 |
Chooses one of six team patterns. By default it produces an agent execution packet; with --run it runs the pattern as an MVP orchestration runtime and records transcripts, cost, token, and cache metrics.
pipeline — sequential handoff: explore → plan → implement → review.fanout — parallel independent investigation, then synthesize.fanin — collect outputs from several agents into one decision.expert-pool — ask specialized reviewers for second opinions.red-team — adversarial review for security/reliability risks.supervisor — one coordinator tracks subtask completion.node .claude/skills/orchestrate/orchestrate.mjs "task description" --pattern=fanout
node .claude/skills/orchestrate/orchestrate.mjs "task description" --pattern=fanout --run --max-concurrency=3
node .claude/skills/orchestrate/orchestrate.mjs "task description" --pattern=fanout --run --transport=mock
node .claude/skills/orchestrate/orchestrate.mjs "task description" --pattern=fanout --run --transport=codex-cli
node .harness/scripts/orchestration-contract-from-task.mjs <task-id>
node .claude/skills/orchestrate/orchestrate.mjs --contract=release-review --run --transport=mock
node .claude/skills/orchestrate/orchestrate.mjs --resume=<run-id-or-dir>
node .claude/skills/orchestrate/orchestrate.mjs --validate-run=<run-id-or-dir>
Packet mode writes .harness/docs/orchestration/<timestamp>-<pattern>.md.
When this skill is rendered for Codex under .agents/skills, runtime mode
defaults to --transport=codex-cli; Claude-rendered skills default to
--transport=claude-cli.
Use orchestration-contract-from-task.mjs <task-id> to derive a contract from
the task contract's risk tier, required reviewers, permissions, and evidence
path before running high-risk or mutating multi-agent work.
Contract mode reads .harness/orchestration/contracts/<id>.json and binds the
run to explicit lanes, tool policies, required reviewers, task ids, and output
artifacts. Use it for high-risk work or any multi-agent run that should be
auditable after the fact.
Runtime mode writes .harness/orchestration/<run-id>/manifest.json, per-agent transcripts, summary.json, and summary.md. It also appends orchestrate telemetry to .harness/telemetry.jsonl so cost/replay/export tools can trace task -> skill -> provider call -> cache bucket -> cost. Use --no-fail-fast when you need every lane to finish even after a failure.
Hardening flags:
--timeout-ms=N kills a stalled lane and records a timeout result.--retries=N retries failed lanes before marking them failed.--resume=<run-id-or-dir> skips previously passed lanes and reruns missing/failed lanes.--cancel=<run-id-or-dir> writes a cancellation marker consumed by resume/running lanes.--validate-run=<run-id-or-dir> checks manifest, summary, and JSONL transcript schema.### Orchestration: <pattern>
### Agents: <count>
### Packet: .harness/docs/orchestration/<timestamp>-<pattern>.md
### Synthesis owner: main agent