rjcorwin/cook/skill

Claude Code and agent skill by rjcorwin/cook

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Orchestrate coding agents with review loops, parallel races, repeat passes, and task-list progression. Use when: the user asks to "cook" something, wants iterative refinement, wants to race multiple approaches, or needs to work through a task list.

Install

npx vskill@latest install rjcorwin/cook/skill

SKILL.md

Cook — Agent Orchestration CLI

cook wraps your coding agent (Claude Code, Codex, OpenCode) in composable workflows: review loops, repeat passes, parallel races, and task-list orchestration.

Important: Never use `--sandbox none`. The default sandbox mode (agent) is correct when running as a skill. It preserves the parent agent's security boundaries.

Quick reference

sh
# Single work call
cook "Implement dark mode"
# Review loop (work → review → gate, repeat until DONE)
cook "Implement dark mode" review
# Repeat 3 times
cook "Improve the design" x3
# Race 3 versions, pick the best
cook "Implement dark mode" v3 pick "cleanest implementation"
# Two approaches, pick the winner
cook "Auth with JWT" vs "Auth with sessions" pick "best security"
# Work through a task list
cook "Do the next task in PLAN.md" \
ralph 5 "DONE if all tasks complete, else NEXT"
# Everything composes
cook "Implement dark mode" review v3 "cleanest result"

Operators

Operators compose left to right. Loop operators wrap everything to their left.

Loop operators
OperatorEffect
reviewAdd a review→gate loop (up to 3 iterations by default)
review NReview loop with up to N iterations
xN / repeat NRun work N times sequentially
ralph N "gate"Outer gate for task-list progression (DONE/NEXT)

Custom review/gate prompts (positional shorthand):

sh
cook "work prompt" "review prompt" "gate prompt"
cook "work prompt" "review prompt" "gate prompt" "iterate prompt" N
Composition operators
OperatorEffect
vN / race NN identical runs in parallel worktrees
vs2+ different runs in parallel worktrees
pick ["criteria"]Resolver: pick one winner (default)
merge ["criteria"]Resolver: synthesize all results
compareResolver: write comparison doc, no merge
Composition examples
sh
cook "A" vs "B" pick "criteria" -y # two approaches, pick winner
cook "A" vs "B" merge "best of both" -y # synthesize both
cook "A" vs "B" compare # comparison doc only (no prompts)
cook "A" v3 "criteria" -y # race 3, implicit pick
cook "A" x3 vs "B" x3 pick "best" -y # per-branch loop operators
Agents must always pass `-y` on composition commands to avoid hanging on prompts.

Flags

-y, --yes Auto-accept all prompts (REQUIRED for agents)
--max-iterations N Max review iterations
--work-agent AGENT Per-step agent override
--review-agent AGENT
--work-model MODEL Per-step model override
--review-model MODEL
--hide-request Hide the templated request panel

IMPORTANT: Always use `-y` when invoking cook as an agent. Composition commands (v3, vs, pick, merge) prompt for confirmation ("Apply Run N?", "Remove worktrees?"). Without -y, the process hangs waiting for interactive input that agents cannot provide. This flag auto-accepts all prompts.

Prerequisites

Before running cook:

  1. The project must have cook init run (creates COOK.md, .cook/config.json)
  2. For composition operators (vs, vN), the working tree must be clean (commit first)

When to use cook vs doing the work directly

Use cook when:

  • The user explicitly asks to "cook" or "let it cook"
  • Multiple iterations of refinement are needed (review loops)
  • Multiple competing approaches should be tried (races, vs)
  • A task list needs sequential progression (ralph)
  • The user wants autonomous completion without manual review cycles

Do the work directly when:

  • It's a simple, one-shot change
  • The user wants to review each step interactively

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
371
Repository forks
12
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/rjcorwin/cook/skill/badge)](https://verified-skill.com/skills/rjcorwin/cook/skill)

Something wrong with this skill or its page? Tell us.