Agent skills and generated runtime entry points for Spec Kit.
This repository tracks the current Spec Kit output together with a small set of repository-specific skills.
- Shared Agent Skills -
skills/contains the sharedspeckit-*skills plus repository-specific skills. Claude Code uses them through.claude/skills, while Codex CLI uses them through.agents/skills. - Cursor Agent - Spec Kit-generated skills live in
.cursor/skills/. - GitHub Copilot CLI - Spec Kit-generated agents and prompt wrappers live in
.github/agents/and.github/prompts/. - OpenCode - Spec Kit-generated commands live in
.opencode/commands/. - Gemini CLI -
.gemini/commands/remains checked in, but Gemini is not part of the current CI regeneration set. - Spec Kit infrastructure -
.specify/contains integration state, manifests, scripts, and templates used by the generated skills and commands.
The tracked Spec Kit version is recorded in VERSION.md. CI checks the latest Spec Kit release and regenerates outputs only when that version changes.
The repository CI currently runs specify init --integration for:
codexagycursor-agentcopilotopencodeclaude
The generated files are therefore not all stored in the same layout:
| Runtime / integration | Repository path | Role |
|---|---|---|
Agent Skills (agy) |
skills/ |
Shared Spec Kit skills |
| Claude Code | .claude/skills -> ../skills |
Shared skills exposed to Claude Code |
| Codex CLI | .agents/skills -> ../skills |
Shared skills exposed to Codex |
| Cursor Agent | .cursor/skills/ |
Cursor-specific generated skill copies |
| GitHub Copilot CLI | .github/agents/, .github/prompts/ |
Generated agents and prompt wrappers |
| OpenCode | .opencode/commands/ |
Generated commands |
| Gemini CLI | .gemini/commands/ |
Checked-in commands; not regenerated by the current CI configuration |
Legacy layouts such as .claude/commands/, .codex/prompts/, and .opencode/command/ are intentionally not maintained.
-
Clone this repository.
git clone https://github.com/dceoy/speckit-agent-skills.git cd speckit-agent-skills -
Install Spec Kit.
-
Initialize a target project with the Spec Kit integration appropriate for your agent.
specify init --here --integration <integration>
-
If the target runtime consumes Agent Skills directly, copy the required skills from
skills/into that project's skills directory when needed.cp -a skills/* /path/to/project/skills/ -
Invoke the relevant
speckit-*skill or runtime command from your agent.
The main Spec-Driven Development flow in this repository is:
- Constitution - Define project principles.
- Specify - Capture requirements for new work.
- Or Baseline - Derive a specification from an existing codebase.
- Clarify (optional) - Resolve material ambiguities.
- Plan - Produce the implementation strategy.
- Tasks - Generate ordered work items.
- Analyze (optional) - Check consistency across the specification, plan, and tasks.
- Converge (optional) - Compare an existing implementation with the spec/plan/tasks and append remaining work to
tasks.md. - Implement - Execute the tasks.
Additional utilities include Checklist for requirements-quality checks and Tasks to Issues for converting generated tasks into GitHub issues.
flowchart TD
C0["speckit-constitution"] --> C1["speckit-specify"]
C0 --> B["speckit-baseline"]
C1 --> C2["speckit-plan"]
B --> C2
C2 --> C3["speckit-tasks"]
C3 --> C4["speckit-implement"]
C1 -.-> O1["speckit-clarify"]
B -.-> O1
O1 -.-> C2
C3 -.-> O2["speckit-analyze"]
O2 -.-> C4
C3 -.-> O3["speckit-converge"]
O3 -.-> C4
C3 -.-> O4["speckit-taskstoissues"]
C1 -.-> O5["speckit-checklist"]
C2 -.-> O5
C3 -.-> O5
The current shared speckit-* set includes:
speckit-analyzespeckit-baselinespeckit-checklistspeckit-clarifyspeckit-constitutionspeckit-convergespeckit-implementspeckit-planspeckit-specifyspeckit-tasksspeckit-taskstoissues
Most Spec Kit-managed skills are tracked by the manifests in .specify/integrations/. speckit-baseline is a repository-specific skill for deriving specifications from existing code.
claude-command-converter- Repository-specific utility for converting Claude command content into Agent Skill format.
.
├── skills/ # Shared and repository-specific Agent Skills
├── .agents/
│ └── skills -> ../skills # Codex CLI access to shared skills
├── .claude/
│ └── skills -> ../skills # Claude Code access to shared skills
├── .cursor/
│ └── skills/ # Cursor Agent generated skills
├── .gemini/
│ └── commands/ # Checked-in Gemini commands
├── .github/
│ ├── agents/ # GitHub Copilot generated agents
│ ├── prompts/ # GitHub Copilot prompt wrappers
│ └── workflows/ # CI and Spec Kit regeneration workflows
├── .opencode/
│ └── commands/ # OpenCode generated commands
├── .specify/
│ ├── integrations/ # Integration manifests
│ ├── scripts/bash/ # Spec Kit helper scripts
│ ├── templates/ # Spec Kit templates
│ ├── init-options.json # Last initialization options
│ └── integration.json # Current integration state
├── .vscode/
│ └── settings.json # Workspace settings
└── VERSION.md # Tracked Spec Kit version
.github/workflows/ci.yml calls .github/workflows/speckit-init.yml. The workflow:
- resolves the latest Spec Kit release,
- records
specify --versioninVERSION.md, - skips regeneration when
VERSION.mdis unchanged, - runs
specify init --force --here --ignore-agent-tools --script sh --integrationfor the configured integrations, - formats generated Markdown and JSON, and
- commits updated Spec Kit outputs when the tracked version changes.
Treat manifest-managed runtime files as generated output. Prefer changing the integration configuration or repository-specific skills instead of manually maintaining generated copies.
Install and authenticate only the runtime tools you intend to use, plus Spec Kit itself.
- Claude Code - Uses
.claude/skills. - OpenAI Codex CLI - Uses
.agents/skills. - Cursor Agent - Uses
.cursor/skills/. - GitHub Copilot CLI - Uses
.github/agents/and.github/prompts/. - OpenCode - Uses
.opencode/commands/. - Gemini CLI - Can use the checked-in
.gemini/commands/, but those files are outside the current regeneration set. - Spec Kit - Install from github.com/github/spec-kit.
- Skills do not always auto-run; use the runtime's skill invocation flow or request the skill explicitly.
- If a skill fails, inspect its
SKILL.mdand verify the required Spec Kit project structure and prerequisites. - Run Spec Kit helper scripts from the repository root and prefer their structured output modes such as
--jsonwhen available. - Do not restore legacy runtime layouts unless a currently supported Spec Kit integration starts generating them again.
See LICENSE for details.