agentafk
Skills

Built-in Skills

Curated overview of the skills shipped with agent-afk and when to reach for each one.

AFK includes a small set of skills out of the box. Each one works as a slash command (type /skillname) and as a tool the model can call on its own.

Public skills

These skills are visible to all users.

/get-started

/get-started

Purpose: Guided first-run onboarding — preflight checks, optional capability setup, and a route to your first task.

What it does: Runs a preflight check (git repo, model provider, AFK.md, Exa/Telegram/service config), asks your name and gives a brief intro, detects importable Claude Code / Codex assets and offers afk migrate, then walks optional capability setup (Exa Search, Telegram via /telegram-setup, background service via /service-setup). Ends by recommending /init to generate project context and /clear to start fresh, then routes you to your first task.

Execution mode: load — runs interactively in the current session (not a fork), so it can drive setup alongside you.

When to use: When you're setting up AFK for the first time or asking how to get going. Triggers on /get-started, "how do I start", "set me up", or "onboard me". Best run in the interactive REPL.


/mint

/mint <idea>
/mint --continue approved

Purpose: End-to-end feature and refactor delivery — from a one-sentence idea to a verified, ship-ready implementation.

What it does: Runs an eight-phase pipeline: spec → research → plan → parallelize → build → verify → heal → ship. After the spec phase completes it pauses and surfaces the spec for your approval. Resume with /mint --continue approved (or yes, lgtm, sure) — the handler reloads state from disk and runs phases 2–8. If the verify phase fails, a heal loop retries up to two times before surfacing a structured failure report.

When to use: When you want a feature or refactor delivered end-to-end without driving each phase manually. Works best on self-contained features with clear acceptance criteria.


/diagnose

/diagnose <bug-or-failing-test>

Purpose: Parallel root-cause analysis for bugs and failing tests.

What it does: Runs parallel agents to analyze your codebase and git history simultaneously. Synthesizes 2–4 hypotheses. Tests each in an isolated git worktree. Returns a validated root cause, a proposed fix, and a named outcome label (clear_winner, multiple_plausible, dissent, all_inconclusive, or no_hypotheses).

When to use: When a test is failing, a bug is reported, or behaviour is unexplained and you need a root cause with evidence rather than a guess.


/service-setup

/service-setup

Purpose: Install an AFK background process (Telegram bot or daemon) as a macOS LaunchAgent so it auto-starts on login and relaunches on crash.

What it does: Runs pre-flight checks (e.g., refuses to install the Telegram service if the token is invalid or missing, which would otherwise cause an infinite KeepAlive crash loop), invokes afk service install, verifies the result with afk service status, and surfaces the management cheatsheet. macOS-only — gracefully refuses on other platforms.

Execution mode: fork — runs as an isolated sub-agent with its own context window.

When to use: When you want afk telegram start or afk daemon to survive reboot, crash, or OOM. Triggers on phrasings like "install as a service", "auto-start on login", "keep the bot running", or "launchd".


/telegram-setup

/telegram-setup

Purpose: Guide you through first-time Telegram bot onboarding without ever putting the bearer token in the model's context.

What it does: Instructs you to run afk telegram setup in a terminal (which handles token entry via stdin), then uses three sanctioned subcommands — afk telegram check-token, afk telegram discover-chat, afk telegram set-allowed-chat <id> — to validate and complete allowlist setup. The token never crosses the LLM provider boundary.

Execution mode: fork — runs as an isolated sub-agent.

When to use: When you want to set up Telegram push notifications for the first time, or when you need to debug a partially-configured install. Also triggers when AFK detects TELEGRAM_BOT_TOKEN is unset and you ask about notifications.


Internal skills

The following skills are registered but hidden from public surfaces unless AFK_INTERNAL=1 is set. They are listed here for reference only.

SkillAudienceNotes
audit-fitinternalAudits ~/.afk artifacts for correct type categorisation; writes migration briefs to ~/.afk/agent-framework/briefs/. Runs fine without the internal tier — it's hidden for UX, not correctness.

Skills registered at runtime by plugins

Plugin skills (~/.afk/plugins/<plugin>/skills/<skill>/SKILL.md) are discovered automatically when the plugin is installed. They surface in /help and tab-complete. See Custom Skills for how to author your own.