steadwing

Steadwing Agent Skills

Official Agent Skills for Steadwing — AI-powered Root Cause Analysis for production incidents.

Install once and your coding agent self-onboards: it registers a Steadwing agent on its own, and is ready to run RCA on errors — no signup, no API key to paste.

Install

Works with Claude Code, Cursor, Windsurf, GitHub Copilot, and 18+ other agents — one repo, every agent. There's nothing to publish to npm and no key to configure: the installer reads this repo directly, and the skill registers your agent with Steadwing the first time it runs.

Any agent (recommended)

npx skills add steadwing/agent-skills

The skills CLI auto-detects your agent and installs the skill into its directory (e.g. .claude/skills/ for Claude Code, .cursor/skills/ for Cursor). Useful flags:

npx skills add steadwing/agent-skills -g                  # global — available in all projects
npx skills add steadwing/agent-skills -a claude-code      # target a specific agent
npx skills add steadwing/agent-skills --skill steadwing   # install just this skill
npx skills add steadwing/agent-skills --copy              # copy instead of symlink

If the skill doesn't appear in Claude Code after install, it's usually a project- vs global-scope mismatch — try -g, or target it explicitly with -a claude-code. Then start a new session (agents load skills at session start).

Claude Code plugin (native)

Claude Code users can also install via the native plugin system, which doesn't depend on the skills CLI:

claude plugin marketplace add steadwing/agent-skills
claude plugin install steadwing@steadwing-agent-skills

The skill activates automatically when relevant — no manual invocation needed.

What you get

Skill	Description
steadwing	Self-registers as a Steadwing agent, triggers AI Root Cause Analysis on error logs/stack traces (attaching relevant source files), optionally installs a hook that watches for production-style errors and offers RCA, and surfaces claim/expiry reminders.

How it works

1. Install the skill with the command above.
2. First use — the skill detects your agent (Claude Code / Cursor / Windsurf), calls POST /api/agents/register (no auth), and stores the returned API key at ~/.steadwing/credentials.json. It prints a claim URL so you can link the agent to your Steadwing account (unclaimed agents expire in 3 days).
3. (Optional) Auto-detection — in Claude Code, the skill can install a PostToolUse hook that detects production-style errors (tracebacks, unhandled exceptions, stack traces) and asks whether to run RCA. The hook only detects and asks — it never sends anything without your confirmation.
4. Run RCA — on request (or after you confirm a detected error), the skill calls POST /api/mcp/analyze with the error log and relevant files, then returns the incident URL to track analysis (~1–3 min, runs in the background).

Repo structure

.
├── .claude-plugin/
│   └── marketplace.json         # Claude Code native plugin manifest
├── skills.sh.json               # skills.sh registry/grouping
└── skills/steadwing/
    ├── SKILL.md                 # agent instructions (entry point)
    ├── references/
    │   └── api-reference.md     # full Steadwing API reference
    └── scripts/
        ├── steadwing_rca_hook.py  # passive PostToolUse error-detection hook
        └── install_hook.py        # idempotent installer for the hook

Privacy

The skill sends error logs and any source files you confirm to the Steadwing API (https://api.dev.steadwing.com) for analysis. RCA is only triggered on explicit request or after you confirm a detected error — never silently.

License

MIT