help

name	help
description	Use when the user asks for help, afk help, what to do next, where am I, which AFK skill to use, AFK workflow orientation, or a broad catalog explanation

Help

Orient the user in the AFK workflow and recommend the next useful step.

Core principle: be specific, not encyclopedic. Ground the answer in observed project state and surface only the skills relevant to the current state or the user's question.

When to Use

Use this skill when the user asks for AFK help, workflow orientation, the next AFK step, where they are in the process, which AFK skill to run, or an explanation of AFK itself.

Sources

• Catalog: afk-help.csv
• AFK familiarity: does brain/ exist? Present = the user has used AFK here before (returning); absent = this is a fresh project (first contact).
• Domain glossary: brain/context.md
• Decisions: brain/decisions/
• Plans: brain/plans/ (from afk:grill and afk:plan)
• External grounding: brain/sources/ (from afk:research)
• Codebase maps: brain/codebase/ (from afk:map-codebase)
• QA evidence: qa/
• Current implementation state: git status --short and git diff --stat

Process

1. Read the catalog.
2. Inspect the available artifacts listed above, including whether brain/ exists (fresh vs returning).
3. If this is a fresh project (no brain/), lead with the onboarding path before any other routing — see "Fresh project" below.
4. Classify the user intent:
   • General orientation: explain where they are and the next step.
   • Specific skill question: explain that skill, when to use it, and what it produces.
   • End-to-end lifecycle request: recommend afk:ship.
   • "What now?": recommend exactly one next skill unless the state is truly ambiguous.
5. Give the invocation, e.g. afk:write-good-goal, afk:grill, afk:ship, afk:implement, afk:batch, afk:simplify, or afk:qa.
6. If one next step is clearly right, offer to run it now.

Routing rules

• Fresh project (no brain/ directory): the user is new to AFK here. Before routing on intent, orient them on the eval-first / TDD habit AFK encourages: write a failing eval first, then implement against it. Recommend afk:write-evals to capture the behavior as a failing eval, then afk:implement to make it pass. Mention afk:init-brain so the memory vault exists for future runs. Skip this lead-in once brain/ is present.
• The user asks for /goal help, objective shaping, success criteria, a completion condition, or a definition of done for long-running agent work: recommend afk:write-good-goal.
• The user asks to run the whole AFK flow, ship a feature, resume an AFK lifecycle, or get from idea/plan to a ship/no-ship verdict: recommend afk:ship.
• The user asks to prototype, mock up, try a few designs, create UI variants, make something playable, or sanity-check a state model/API/data shape by interacting with it: recommend afk:prototype.
• The user asks for prior art, external best practice, how other teams or libraries solve a problem, competitor patterns, or a scan of the landscape before planning: recommend afk:research to turn the open question into an external-grounding digest. It feeds afk:grill.
• No concrete plan in brain/plans/: recommend afk:grill to interview a vague idea into a plan, or afk:plan to decompose an already-clear task into phased plans. Both feed afk:implement.
• A plan exists and there is little or no implementation diff: recommend afk:implement. If that plan splits into many independently-mergeable units (a codebase-wide migration, rename, or repeated pattern change) and the user wants parallel PRs, recommend afk:batch instead.
• There is an implementation diff after a plan: recommend afk:simplify.
• The user asks for verification, browser QA, API/service QA, screenshots, request/response evidence, or a ship/no-ship call after implementation: recommend afk:qa. It is the closer for the afk:implement path; afk:batch opens its own PRs, and afk:ship stops at the QA verdict.
• The user asks about persistent memory, project principles, capturing learnings, or the brain/ vault: recommend the memory skills — afk:reflect to capture a session, to mine past conversations, to audit and prune, for direct read/write, to scaffold, to map an existing area as-is, for phased plans, and for a principle-grounded review. These run anytime, not as a fixed flow phase.

Missing Artifacts

If a file or directory is missing, treat that as workflow signal. Do not invent project-specific progress, decisions, plans, implementation, or QA evidence.

When the catalog is missing or unreadable, STOP and say that the help catalog is unavailable. Give only the recommendation that can be justified from the observed project state.

If AFK skills don't auto-trigger

AFK relies on Claude Code injecting each skill's description so it fires on intent. On some Claude Code versions plugin skills are not auto-discovered, or a crowded install drops less-used descriptions from the skill listing. When a skill does not trigger on its own, invoke it explicitly:

/afk:help, /afk:write-good-goal, /afk:prototype, /afk:grill, /afk:implement, /afk:batch, /afk:simplify, /afk:qa, /afk:ship.

If auto-trigger keeps failing, run /doctor to check plugin/skill loading, and raise skillListingBudgetFraction in settings so more descriptions stay listed.

Stop and Ask

STOP and ask a clarifying question when two or more next skills are equally plausible and the project state does not break the tie.

Ask only for the missing intent or source of truth needed to choose the next AFK step.

Red Flags

Thought	Reality
"The user asked for help, so list every skill."	List only relevant skills unless they ask to show everything.
"The plan probably exists somewhere."	Inspect brain/plans/; missing artifacts are workflow signal.
"There is a diff, so QA is next."	After a plan plus implementation diff, recommend afk:simplify before QA unless the user explicitly asks for verification.
"The user said ship, so recommend QA only."	afk:ship owns the full route to a verdict; QA is just the final evidence phase.
"Just route on intent; familiarity doesn't matter."	No brain/ means a fresh project — lead with the eval-first onboarding (afk:write-evals then afk:implement) before intent routing.

Output

Use this compact shape:

• Where you are: one sentence grounded in observed files or git state.
• Next step: [menu-code] Display Name and skill invocation.
• Why: one sentence.
• Expected output: the artifact or state the skill should produce.

When the user asks a broad catalog question, list only the relevant skills. When the user asks "show me everything", show the full catalog grouped by phase.