Cook Skill

End-to-end feature implementation with smart intent detection.

Installation

Copy the cook/ folder to your Claude skills directory:

cp -r cook ~/.opencode/skills/

Usage

/ck:cook <natural language task OR plan path>

The skill automatically detects your intent and routes to the appropriate workflow.

Examples

# Interactive mode (default)
/ck:cook implement user authentication

# Execute existing plan
/ck:cook plans/260120-auth

# Fast mode (skip research)
/ck:cook quick fix for login bug
/ck:cook implement feature --fast

# Auto mode (trust me bro)
/ck:cook implement dashboard trust me
/ck:cook implement feature --auto

# Parallel mode (multi-agent)
/ck:cook implement auth, payments, notifications
/ck:cook implement feature --parallel

# No-test mode
/ck:cook implement feature --no-test

Modes

Mode	Research	Testing	Review	Use Case
interactive	✓	✓	User approval	Default, full control
auto	✓	✓	Auto if score≥9.5	Trusted, hands-off
fast	✗	✓	Simplified	Quick fixes
parallel	Optional	✓	User approval	Multi-feature work
no-test	✓	✗	User approval	Speed priority
code	✗	✓	User approval	Existing plans

Intent Detection

The skill detects mode from:

Explicit flags: --fast, --auto, --parallel, --no-test
Plan paths: ./plans/*, plan.md, phase-*.md
Keywords: "fast", "quick", "trust me", "auto", "no test"
Feature count: 3+ features → parallel mode

Workflow

[Intent Detection] → [Research?] → [Plan] → [Implement] → [Test?] → [Review] → [Finalize]

Files

cook/
├── SKILL.md                           # Main skill definition
├── README.md                          # This file
└── references/
    ├── intent-detection.md            # Detection rules
    ├── workflow-steps.md              # Step definitions
    ├── review-cycle.md                # Review process
    └── subagent-patterns.md           # Subagent usage

Version

2.1.0 - Review gates added for human-in-the-loop mode 2.0.0 - Smart intent detection (hybrid approach)

2.2 KiB Raw Blame History