Agent skills for agentsfleet — install, drive, and operate agents from any AI coding host.

Long-lived runtimes that own one operational outcome end to end. These skills teach Claude Code, Codex CLI, Amp, and OpenCode to drive agentsfleet non-interactively so your agent can install, steer, and inspect agents without you reading every flag.

---

Installation

Skills here follow the Agent Skills format and work with 18+ AI agent hosts including Claude Code, Codex CLI, Amp, OpenCode, and Cursor.

Install all agentsfleet skills

npx skills add agentsfleet/skills

This symlinks every top-level agentsfleet-* directory into each supported host's skill path that exists on your machine (~/.claude/skills/, ~/.codex/skills/, ~/.amp/skills/, ~/.opencode/skills/).

Install a single skill

npx skills add agentsfleet/skills --skill agentsfleet-install-platform-ops

Claude Code Plugin

These skills also ship as a Claude Code plugin marketplace:

# 1. Register the marketplace
claude plugin marketplace add agentsfleet/skills

# 2. Install the plugin you want
claude plugin install agentsfleet-install-platform-ops@agentsfleet-skills

Prerequisites

Most skills here drive agentsfleet. Install it first:

npm install -g @agentsfleet/cli
agentsfleet auth login

Available Skills

agentsfleet-install-platform-ops

One-command install of the platform-ops agent on a user's repo. Watches GitHub Actions CD failures and posts evidenced diagnoses to Slack.

Use when:

• Setting up an agent on a new repo for the first time
• Resolving tool credentials (Fly, Slack, GitHub, Upstash) via 1Password / env / prompt
• Registering and HMAC-verifying webhooks from the user's local gh
• Smoke-testing the install with a real steer round-trip

Slash-command: /agentsfleet-install-platform-ops

Usage

Skills are automatically available once installed. Invoke them by their slash-command in any supported host:

/agentsfleet-install-platform-ops

The agent reads the skill body, walks the install plan, and surfaces every failure mode verbatim so you can resolve it before retrying.

Skill Structure

Each skill is a top-level directory matching its slash-command:

agentsfleet-install-platform-ops/
├── SKILL.md              # Frontmatter + body the host LLM reads
├── references/           # Detailed docs the skill loads on demand
│   ├── credential-resolution.md
│   ├── failure-modes.md
│   └── self-managed-handoff.md
└── evals/                # Pin-tests + LLM-judge harness (bun test)
    ├── package.json
    ├── tsconfig.json
    ├── skill-body.test.ts
    ├── skill-runtime.test.ts
    ├── repo-detection.test.ts
    ├── llm-judge.eval.ts
    └── fixtures/         # Synthetic repos for runtime assertions

The SKILL.md frontmatter follows the Agent Skills Open Standard with name, description, license, metadata, inputs, and references.

Release model

main is the release surface. Push to main = ship. No tags, no semver — same model as a dotfiles repo. npx skills add agentsfleet/skills always pulls the latest commit.

Contributing

Pull requests welcome. Each skill change runs evals on PR via .github/workflows/eval.yml (Bun test runner). New skills should add a top-level <skill-name>/ directory with SKILL.md, references/, and evals/, plus a matching entry in .claude-plugin/marketplace.json.

License

MIT — Copyright (c) 2026 agentsfleet