writing-skills
Use when creating new skills, editing existing skills, verifying skills work before deployment, or when creating/editing any document an agent must read (`.devin/global_rules.md`, `.devin/rules/*.md`, docs reached by pointer).
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- writing-skills
- triggers
- [ "user", "model" ]
- description
- Use when creating new skills, editing existing skills, verifying skills work before deployment, or when creating/editing any document an agent must read (`.devin/global_rules.md`, `.devin/rules/*.md`, docs reached by pointer).
02 资源文件 13
- 参考文档examples/AGENTS_MD_TESTING.md5.4 KB
- 其他graphviz-conventions.dot5.8 KB
- 参考文档modes/creation.md25.5 KB
- 参考文档persuasion-principles.md5.8 KB
- 参考文档reference/examples/AGENTS_MD_TESTING.md5.4 KB
- 其他reference/graphviz-conventions.dot5.8 KB
- 参考文档reference/persuasion-principles.md5.8 KB
- 脚本reference/render-graphs.js4.7 KB
- 参考文档reference/SKILL-MECHANICS.md2.6 KB
- 参考文档reference/testing-skills-with-subagents.md12.2 KB
- 参考文档reference/writing-for-agents.md10.8 KB
- 脚本render-graphs.js4.7 KB
- 参考文档testing-skills-with-subagents.md12.2 KB
03 SKILL.md 正文
Writing Skills
Writing skills IS TDD applied to process documentation. Write test cases (pressure scenarios with subagents), watch them fail (baseline), write the skill, watch agents comply, refactor (close loopholes). If you didn't watch an agent fail without the skill, you don't know it teaches the right thing.
Full creation workflow: modes/creation.md. Agent-facing doc principles:
reference/writing-for-agents.md. Frontmatter/invocation mechanics:
reference/SKILL-MECHANICS.md. Subagent pressure-testing:
reference/testing-skills-with-subagents.md. Persuasion:
reference/persuasion-principles.md. Graphviz conventions + renderer:
reference/graphviz-conventions.dot, reference/render-graphs.js.
Examples: reference/examples/.
Essentials
- Personal skills:
~/.config/devin/skills/<name>/(%APPDATA%\devin\skills\on Windows). Project skills:.devin/skills/<name>/..agents/skills/is the cross-runtime standard. - A skill is a reference guide for proven techniques — not a narrative of how you solved a problem once.
- Frontmatter:
name+descriptionwritten as discovery triggers ("Use when…"), not workflow summaries. Optional:allowed-tools,subagent,agent,model(free variants only),permissions,triggers. - Keep SKILL.md small (~<10KB); push detail behind pointers
(
reference/*.md) — progressive disclosure protects the hierarchy.
Context pointers (the triggering layer)
A pointer's wording decides when material gets reached. Front-load the leading word; one trigger per branch (synonyms = one branch twice); cut identity the body carries. Always-loaded pointer text spends context every turn — prune harder than the body.
The two loads
- Context load — always-loaded material costs tokens/attention every turn.
- Cognitive load — the human index of which docs exist. Spend where human judgement matters; remove elsewhere.
Information hierarchy
- In-file steps (primary) → 2. in-file reference → 3. disclosed reference (separate file behind a pointer). Inline what every branch needs; disclose what only some branches reach.
Rehearsal gate (visual artifacts)
Skills that produce visual output (animations, mockups, UI stills) verify by artifact, not only by tests. Declare before building:
- a deterministic fault checklist the deliverable must pass (per-frame contact sheet, loop seam, target widths, contrast rules);
- a mandatory handover artifact naming the source example copied, what changed, and what was NOT checked;
- a fresh-eye review pass before declaring done.
Prior art: stageshow in benjitaylor/skills (external install via
npx skills add; unvendored - no license). Issue #428.
Quality checklist (before commit)
name/descriptionper spec; discovery keywords in description.- Devin-native tools and paths only; no non-Devin platform leakage.
- Subagent profiles per Rule 20 (
researcher, neversubagent_explore). - Python for cross-platform helper scripts.
- Visual-output skills declare their artifact gate (fault checklist + handover doc + fresh-eye pass).
- No AI signatures.
- Pressure-tested with subagents where compliance matters.
04 同仓库更多 Skills
montador-de-ideias
Use when assembling, expanding, or elevating a raw project idea or initial prompt into an ultra-detailed, first-principles dossier and compiled master prompt with journalistic OSINT market research, Tech Lead and Staff Architect justifications, atomic and modular hardening/optimization, and high-impact contact surfaces.
git-github-adaptive-engineering
Use when any task touches git history, GitHub issues/PRs/releases/CI, repo governance, or branch and merge decisions. Runs the Fase Zero situational-awareness battery, picks one of five delivery flows via the adaptive decision engine, and enforces commit/PR information-density standards. Orchestrates the gh, git-workflows, github-actions, pr, and debug/finish skills.
adaptive-agenda-calendar-routine-engine
Use when planning or executing work across hours, days, weeks, or sprints. Turns the calendar into a strict time-blocked agenda (HH:MM-HH:MM blocks carrying a routine profile, expected inputs, and a handoff output artifact), runs the 5-stage cognitive flow (Radar and Hunt -> Triage plus Steve Jobs conception -> workbench manifest -> >=32-cycle loop -> Close plus Open-Minded Scrum), and unfolds macro calendar milestones into micro agenda blocks through extensions/gcal/wrapper.py.
demo
Use when demonstrating a skill in a fixture project.
duplicate-skill
Use when testing duplicate skill name detection.
duplicate-skill
Use when testing duplicate skill name detection.