Reusable AI agent skills maintained in this repository. The work is shaped by
Sebastian Wessel's article
Spec-Driven Development,
which explains this repository's spec-first approach for keeping autonomous AI
agents aligned from business intent through implementation and review. Each
skill lives in its own subdirectory under skills/ and follows Anthropic's
Agent Skills best practices.
Contents
- Skill Index
- Spec-Driven Workflow
- Spec-Driven Development
- Installation
- Repository Layout
- Documentation Pattern
- Adding a Skill
- Maintainer Notes
Skill Index
| Skill | Description | Human docs |
|---|---|---|
| agent-skill-architect | Creates, reviews, repairs, and optimizes agent skills. Use when a skill needs authoring, trigger tuning, evals, checks, packaging readiness, or drift controls. | docs |
| spec-architect | Creates and updates canonical implementation specs. Use when specs need authoring, evolution, source-of-truth repair, contracts, UX, NFRs, release, or supply-chain definition before readiness review. | docs |
| spec-driven-workflow | Coordinates the spec-driven lifecycle. Use when choosing order across specs, plans, tickets, reviews, feedback routing, or pause/resume. | docs |
| spec-implementation-planner | Turns approved specs into waves, dependency indexes, AFK tickets, and status tracking. Use when specs need parallel-agent planning, ticket readiness, blockers/unblocks, or plan gap checks. | docs |
| spec-implementation-review | Reviews implementation waves against approved specs and tickets. Use when completed or partial work needs acceptance, merge, release, or handoff review with path tracing and persisted findings. | docs |
| spec-readiness-review | Reviews, repairs, and approves implementation spec readiness. Use when specs need semantic review, deterministic checks, readiness reports, approval gates, or gap repair before planning. | docs |
| spec-ticket-implementation | Implements exactly one approved spec plan ticket in fixed scope. Use when a ready ticket needs code changes with contract/codegen-first, test-first, verification, and no invention. | docs |
Spec-Driven Workflow
Use spec-driven-workflow when a task
spans more than one lifecycle stage or when the next step is unclear.
- Specify with
spec-architectuntil specs are approved. - Plan with
spec-implementation-planneruntil waves, tickets, dependencies, scopes, status, and test-first order are ready. - Implement one ready ticket with
spec-ticket-implementation. - Review ticket sets or waves with
spec-implementation-review. - Route findings back to the owning skill and repeat until no blocking findings remain.
Spec-Driven Development
Read Spec-Driven Development for the rationale behind this repository. The article describes how clear specs, implementation plans, ticket execution, and review gates work together so AI agents can build production-ready systems without drifting from the intended business outcome.
Installation
Install this skill package with the open skills ecosystem CLI:
npx skills add sebastianwessel/skills
See Vercel's announcement of
skills
and skills.sh for the package ecosystem.
Repository Layout
skills/
└── <skill-name>/
└── SKILL.md
Optional per-skill directories include references/, scripts/, assets/,
and agents/.
Human-facing skill documentation lives outside the executable skill package:
docs/
└── skills/
└── <skill-name>.md
Documentation Pattern
Each skill should have a concise human docs page with the same structure:
- What the skill is for.
- Installation.
- What it does.
- How it works.
- Workflow or modes.
- Included files.
- Validation and safety notes.
Adding a Skill
-
Create
skills/<skill-name>/SKILL.md. -
Use concise YAML frontmatter with
nameanddescription. -
Keep detailed reference material in files linked directly from
SKILL.md. -
Add
docs/skills/<skill-name>.mdfor human-facing documentation. -
Refresh this table of contents:
python3 scripts/update-readme.py
Maintainer Notes
Repository instructions for agents are in AGENTS.md. Claude reads CLAUDE.md, which points back to the same shared guidance.