patterns
Shared workflow patterns referenced across orchestrator skills — multi-explorer deployment, the discovery/friction signal vocabulary, the outcome interview, and state-chain file formats. Load on-demand when a skill points to a section.
- License 许可
- MIT
- 最近更新
- 2026-11-10
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- patterns
- description
- Shared workflow patterns referenced across orchestrator skills — multi-explorer deployment, the discovery/friction signal vocabulary, the outcome interview, and state-chain file formats. Load on-demand when a skill points to a section.
02 资源文件 5
- 参考文档agents/openai.yaml318 B
- 参考文档discovery.md2.5 KB
- 参考文档explorer.md2.9 KB
- 参考文档interview.md2.3 KB
- 参考文档state-files.md2.1 KB
03 SKILL.md 正文
Shared Patterns
Reusable patterns extracted from the orchestrator skills so they live in one place. Load the specific pattern a skill points you to — don't preload.
Available Patterns
| Pattern | Depth file | Referenced by |
|---|---|---|
| Multi-Explorer | explorer.md |
work, verify |
| Discovery Signals | discovery.md |
verify, capture |
| Friction Signals | discovery.md |
verify, capture |
| Interview Protocol | interview.md |
work |
| State Files | state-files.md |
work, verify, capture |
When to Load
A skill says "see the patterns skill § X" or "see patterns/explorer.md". Read the matching file only when you hit that reference. The sections below are the quick reference; the depth files carry formats, examples, and edge cases.
Multi-Explorer Pattern
Deploy read-only Explore sub-agents to gather context before acting. Each gets one viewpoint over a fixed scope and returns bounded, evidence-backed findings; you cross-reference them and verify the cited lines before trusting anything.
Task(subagent_type="Explore", prompt="[Viewpoint] for [SCOPE]:
- <focused questions for this viewpoint>
Return findings under 2000 characters, each with file:line and (for audits) a severity.")
Explorers cannot load skills — after they report, YOU invoke any matching best-practices skill and reconcile. Full deployment guidance, the viewpoint menu, and the validation checklist: explorer.md.
Discovery Signals
Emit when you find something outside the current scope and want it tracked without derailing. The stop action stop/actions/discovery-extraction.sh scans session output for [DISCOVERY:*] and appends each to context/DISCOVERIES.md; the capture skill later turns those into issues.
| Signal | Meaning | Action |
|---|---|---|
[DISCOVERY:BLOCKER] |
Security/requirements wrong; blocks current work | STOP, escalate |
[DISCOVERY:DEFECT] |
Bug in existing code | Log, continue |
[DISCOVERY:ADJACENT] |
Related to this epic, a different phase | Log for later phase |
[DISCOVERY:DISTANT] |
Unrelated to current work | Log for backlog |
Format, examples, and the triage matrix: discovery.md.
Friction Signals
Emit when the process itself was wrong — a skill, hook, or workflow step got in the way. The stop action stop/actions/friction-extraction.sh scans for [FRICTION:*] and appends each to context/FRICTION_LOG.md; capture classifies each PROJECT vs FRAMEWORK.
| Signal | When |
|---|---|
[FRICTION:SKILL] |
A skill was wrong, outdated, or missing for this codebase |
[FRICTION:HOOK] |
A hook blocked or fired when it shouldn't have |
[FRICTION:WORKFLOW] |
A workflow step was unnecessary or in the wrong order |
[FRICTION:CONTEXT] |
Missing context caused confusion or rework |
Format and the classification handoff: discovery.md.
Interview Protocol
Use the AskUserQuestion tool for iterative Q&A before writing a spec or making a UX/architecture decision the code can't answer. Ask 1-4 structured questions per round, process answers, ask again — never dump a wall of text. Skip for single-file fixes and clear technical tasks. Categories, rhythm, and output: interview.md.
State Files
The state chain carries work across sessions and context resets.
| File | Purpose |
|---|---|
docs/CURRENT_WORK.md |
Active epic/sprint/phase pointer |
context/WORKING.md |
Append-only session observation blocks |
Phase PROGRESS.md |
Implementation checklist |
Read order and templates: state-files.md.
04 同仓库更多 Skills
ponytail
Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra. Use on ANY coding task: writing, adding, refactoring, fixing, reviewing, or designing code, and choosing libraries or dependencies. Also use whenever the user says "ponytail", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", or complains about over-engineering, bloat, boilerplate, or unnecessary dependencies. Do NOT use for non-coding requests (general knowledge, prose, translation, summaries, recipes).
workflow/skill
Patterns for adding or editing a skill — frontmatter contract, router + sub-file split, line budgets.
workflow
Meta-skill for editing the framework itself — skills, hooks, commands, rules, CLAUDE.md. Use before changing anything under .claude/ or the routing docs, so the change keeps the system's shape (lazy-load, forge markers, hook auto-discovery, harness).
work
Work lifecycle orchestration: start, continue, pause, handoff, done, plus epic/sprint planning. Context gathering + goal-focused state-chain transitions.
verify
Verification and auditing. Validate acceptance scenarios, run multi-perspective audits, check type-check/build evidence, and gate phase/sprint completion.
team-lead
Coordinate an Agent Team for parallelizable work — implementation, audit, research, or testing. Teaches the mechanical constraints (WAIT pattern, file ownership, model tiers) and lets you design the team. For orchestration_tier=cluster / multi-agent runs.