agentleFS
doivamongdoivamong/agent-workbench/.claude/skills/awb-cook/SKILL.md

awb-cook

Skill2 starsChanged 6 months agoMITView on GitHub

WHAT: an advanced build workflow that adds three things to a plain plan-then-code flow — graduated oversight modes (how often it stops to ask you), multi-perspective planning (several independent plans, merged), and explicit orchestration of the guard skills at the right steps. USE WHEN: a larger / higher-stakes / multi-module change where a single plan would hide blind spots, or you want to dial the human-checkpoint frequency and plan from several independent angles ("cook this", "run the full workflow"). DO NOT TRIGGER: a routine single-file change (use awb-plan-then-code); a one-line fix; a pure question; work where you do not want sub-agent fan-out. It enforces nothing — it is a bypassable exemplar, not a gate.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    HARD GATE (interactive/fast): approve the merged plan: before code.

    line 50
  2. Distinctive:

    When you fan out a verification or adversarial pass to a sub-agent, hand it the artifact and the contract it must meet — never your own …

    line 57

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Cook (advanced orchestration workflow)

Announce on activation: "Using awb-cook — pick an oversight mode, plan from several angles, then build through the guard skills."

For an ordinary change, awb-plan-then-code is the right, lighter tool — cook is its heavier sibling for larger/riskier work, not a replacement.

Scope

  • Does: orchestrate plan → implement → verify → review while letting you set the checkpoint frequency and synthesizing multiple independent plans.
  • Does NOT: enforce anything (every gate here is advisory — only a human or a real hook makes a gate binding), replace the guard skills it calls, or guarantee the merged plan is correct.

Oversight modes (pick one at the start)

ModeStops to ask you atUse when
interactive (default)the plan, and before commitnormal work
fastonly at an ambiguity or a destructive stepyou trust the scope and want flow
autonothing (reports at the end)a well-specified, low-risk, reversible task

Looser oversight trades safety for speed; reserve auto for reversible, fully-specified tasks.

Process

  1. Frame the task — restate goal + acceptance criteria. If vague, refine first (prompt-refiner).
  2. Multi-perspective plan. Produce several independent plans from different angles (MVP-first, risk-first, reuse-first) — ideally as parallel sub-agents so they don't anchor on each other (see docs/orchestration.md). Then merge: keep what they converge on, surface where they diverge (a real decision), fold in any unique insight. A single plan hides its own blind spots — the merge is the point.
  3. HARD GATE (interactive/fast): approve the merged plan before code.
  4. Implement in small steps; hand non-trivial logic to awb-tdd.
  5. Run the guards at their step (don't reinvent them): long generation → awb-output-guard; config access → awb-config-guard; before commit → awb-review.
  6. Verify with evidence — run the tests/demo and read the output; never report done on a guess. When you fan out a verification or adversarial pass to a sub-agent, hand it the artifact and the contract it must meet — never your own conclusion: a verifier given your claim tends to rubber-stamp it (and sub-agents inherit your framing). Watch for doubt-theater — if across two or more passes the reviewer keeps raising points yet none change the decision, you are validating, not verifying; stop and escalate the call rather than loop a third time.
  7. HARD GATE (interactive): confirm before commit. In fast/auto, commit per the mode and report.

Honesty / limits

Cook enforces nothing — it orchestrates other skills; only a human or a real hook makes a gate binding. The multi-perspective merge improves a plan, it does not prove it correct. The graduated-oversight and multi-perspective-planning patterns are re-authored from github/spec-kit (MIT) and anthropics/claude-plugins-official (MIT); the artifact-not-claim adversarial-verify framing is re-authored from addyosmani/agent-skills (MIT) doubt-driven-development — see THIRD_PARTY_NOTICES.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.