Skip to content
Back to skills

Afk

ASecurity

afk: Part of the afk pipeline. Execute an operator-supplied, pre-reviewed issue/PR scope autonomously through the full pipeline. Requires explicit scope; never pick tracker work. Triggers include \"/afk\", \"AFK mode\", \"go AFK on …\".

  • 2 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added September 26, 2026
ai-agentsgo

Security analysis

A100/100

Pro scans all 12 files and shows the line behind each finding

Scanned September 29, 2026

npx -y skills add AlvinShenSSW/afk --skill afk --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Afk?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Afk
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/alvinshenssw-afk/badge)](https://www.skillsdirectory.com/skills/alvinshenssw-afk)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: afk
description: "afk: Part of the afk pipeline. Execute an operator-supplied, pre-reviewed issue/PR scope autonomously through the full pipeline. Requires explicit scope; never pick tracker work. Triggers include \"/afk\", \"AFK mode\", \"go AFK on …\"."
---

# afk

Hand-off mode: the operator designed and reviewed a scope; you execute exactly
that queue autonomously and stop yourself when done or stuck. This file is the
driver contract; read each reference at the stage where it applies.

## Required reading by stage

Before a dependent action, read the listed reference and its applicable routes.
Reuse a reference only if that same installed revision has already been read
and remains available in the current context. Otherwise reread it. A missing
required reference is an unavailable prerequisite for the dependent action,
not permission to act as though its rules were loaded; retain independent work.
These are level 3 workflow instructions, not an automatic loader.

| Stage | Read before proceeding |
| --- | --- |
| Kickoff/config resolution | [Environment](references/environment.md), [kickoff](references/kickoff.md), [external role profile](references/external-review.md), [issue allowance](references/review-convergence.md), and [CI mode](references/publication.md). |
| Run claim, stage/session handoff or resume | [Continuity](references/continuity.md); keep the current run and authority, including restricted executor handoffs. |
| Spawning a nested child stage | [Delegation](references/delegation.md); supply the bounded task and retained run, then validate the returned result. |
| Stage result or execution summary | [Stage output](references/output.md); retain complete source evidence. |
| Explicit direction-state initialization, check or amendment | [Direction state](references/direction-state.md); no automatic audit activation or dispatch. |
| Retained direction policy enables a stage/signal audit or endpoint check | [Direction audits](references/direction-audit.md); preserve off/shadow/required behavior and the selected endpoint. |
| Plan | [Planner](../afk-spec-planner/SKILL.md); obtain its frozen issue contract before implementation. |
| Design debate or optional design gate | [Design review](references/design-review.md) and [common convergence](references/review-convergence.md). |
| Implementation/self-review | [Implementation pilot](../afk-implementation-pilot/SKILL.md) and [common convergence](references/review-convergence.md). |
| Internal review | [Internal review](../afk-internal-review/SKILL.md) and [common convergence](references/review-convergence.md). |
| External role | [External review](references/external-review.md), [common convergence](references/review-convergence.md), and the selected gate below. |
| Context, receipts or receipt reuse | [Review evidence](references/review-evidence.md) before supplying or reusing those artifacts. |
| Publication, local completion or final report | [Publication](references/publication.md) and [continuity](references/continuity.md). |

External gate entry points: [afk-codex-review](../afk-codex-review/SKILL.md),
[afk-claude-review](../afk-claude-review/SKILL.md),
[afk-kimi-review](../afk-kimi-review/SKILL.md),
[afk-glm-review](../afk-glm-review/SKILL.md),
[afk-deepseek-review](../afk-deepseek-review/SKILL.md), and
[afk-mimo-review](../afk-mimo-review/SKILL.md).

## Per issue — the full waterfall (one at a time)

A nested child's completion is not queue completion. Receive its bounded result,
checkpoint the existing run records and continue the next authorized stage below;
an outstanding result preserves independent queued work. Standalone satellite
requests retain their own endpoints and do not start this waterfall.

**Every issue runs the full review waterfall.** With `remote-ci: off`, the
endpoint is local completion ([Remote checks](references/publication.md#remote-checks)); otherwise each in-scope PR passes
internal review AND the external gate(s) AND lands green (merged, or under
`leave-open` declared ready only after internal review + gate + full test
suite + the revision's check reading). A design doc, a pushed branch, or a PR
not yet AFK merge-ready is a mid-waterfall
checkpoint — never a stopping point and never an operator handoff. "Next:
operator runs the review" is a bug, not an end state.

design doc with a frozen issue contract → adversarial debate ([design-review rules](references/design-review.md);
the shared review-cycle allowance governs repairs) →
design-stage external gate (opt-in pilot, default off; one role per evaluation —
[external design review](references/design-review.md#external-design-review)) → tests
first (targeted) → implementation → adversarial sweep →
commit → push early and open a Draft PR (unless `remote-ci: off`) →
**internal review** (`afk-internal-review`) → triage
every finding and
batch-fix admitted P1s with only inseparable corrections →
**external gate(s)** ([loop, closure, and termination](references/review-convergence.md)) →
**full test suite once** (the project's test command from `.afk/config.md`) on
the final commit → local completion if `off`, otherwise mark Ready for review
to start CI → resolve remote checks → declare AFK merge-ready → merge per policy. The design doc matters more
than the code.

- Scale design/debate depth to the work: mechanical, well-specified work gets a
  brief design and one debate round; design-heavy work gets the full treatment.
  Never scale down tests or gates.
- **Local completion** requires all configured independent roles and the final local suite,
  with the same finding dispositions as the review waterfall. It is not CI approval.
- **Green** = the full test suite green on the final commit, and that commit's
  check reading resolved ([Remote checks](references/publication.md#remote-checks)). A green status on the PR alone is
  not green. Never declare it ready before the suite is green.

Before implementation, freeze the issue contract defined by the
[planner](../afk-spec-planner/SKILL.md). Map implementation edits and admitted
findings to it. Repository evidence may correct it; reviewer preference cannot
expand it. Record other proposals as `OUT-OF-SCOPE`, without implementation or
automatic follow-up issue creation.

## Quality and proportionality

Complete every accepted requirement with appropriate context, capable models,
relevant failure-path and regression checks, and the configured independent
reviews. Do not invent token, call, cost or whole-task time ceilings merely to
save money. Explicit operator limits still apply; an unset limit is not zero.
Retain isolation, credential, transport and cancellation controls for their actual
technical purpose, with enough capacity for complete evidence.

Every abstraction, dependency or additional workflow needs a concrete accepted
requirement or demonstrated defect. Prefer existing mechanisms and the smallest
complete causal correction; simpler code may still require thorough verification.
Use the existing root-cause checkpoint for repeated non-progress instead of
creating new supervisory infrastructure.

## Autonomy

Decide with best-practice defaults and record each decision; do not block on
in-scope work. In-scope risky changes use a justified safe direction; uncertainty alone never
authorizes a new guard or fallback. Only stop for: out-of-scope work, a destructive or
outward-facing action without authorization, or genuine ambiguity with no safe
default. Never merge a PR that is not green or has an open finding — open
meaning `UNTRIAGED`, `Contested`, or an admitted P1 without a closing
disposition. Deferred minor/out-of-scope notes do not bar merge; a structural P2
uses the operator-owned auto-merge bar in [publication](references/publication.md#merge-bar).
Never touch
another session's branch; never deploy (merge ≠ deploy).

## Supported review context

Before supplying review context, read the
[canonical context guidance](references/review-evidence.md#supported-review-context).

## Canonical review receipts

Before supplying or reusing receipts, read the
[canonical receipt guidance](references/review-evidence.md#canonical-review-receipts).

Files in this skill

  • SKILL.md7.9 KB
  • references/continuity.md8.8 KB
  • references/design-review.md10.2 KB
  • references/direction-audit.md9.5 KB
  • references/direction-state.md5.5 KB
  • references/environment.md2.5 KB
  • references/external-review.md10.6 KB
  • references/kickoff.md3.3 KB
  • references/output.md3.5 KB
  • references/publication.md7.2 KB
  • references/review-convergence.md13.3 KB
  • references/review-evidence.md4.8 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.