AgentSkillsCN

coder:clone-behavior

通过提取行为并从头开始重新实现，来克隆一个程序

SKILL.md
--- frontmatter
name: coder:clone-behavior
description: Clone a program by extracting behaviors and reimplementing from scratch
argument-hint: "<source-path> <new-project-name> [--language rust] [--framework candle]"
disable-model-invocation: true
allowed-tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - Bash
  - Task
  - TodoWrite
  - AskUserQuestion

Coder Clone-Behavior

Clone an entire program by extracting WHAT it does (behavior), not HOW it's coded. Then reimplement from scratch - same functionality, completely different code.

Dynamic Context

CLI analysis: !python3 -m erirpg.commands.clone_behavior $ARGUMENTS --json 2>/dev/null || echo '{"error": "CLI failed or dry-run needed"}'

Clone state (if resuming): !cat clone-state.json 2>/dev/null || echo '{"status": "not_started"}'

CLI Options

bash
python3 -m erirpg.commands.clone_behavior <source-path> <new-project-name> [options] --json
OptionDescription
--language <lang>Target language (default: same as source)
--framework <framework>Target framework
--skip-testsDon't extract test contracts
--dry-runShow plan without executing
--modules <list>Only clone specific modules (comma-separated)
--exclude <list>Skip specific modules (comma-separated)

The 5 Phases

code
SOURCE ──► SCAN ──► PLAN ──► IMPLEMENT ──► VERIFY ──► COMPLETE ──► TARGET
         (extract)  (roadmap)  (build)     (diff)    (finalize)

Phase 1: SCAN - Extract Behaviors

  1. •Map source codebase with eri-codebase-mapper
  2. •Extract BEHAVIOR.md for each module with eri-behavior-extractor
  3. •Verify all behaviors extracted

Parallel execution: Up to 4 behavior extraction agents.

See reference.md for details.

Phase 2: PLAN - Create Roadmap

  1. •Initialize target project directory
  2. •Copy BEHAVIOR.md files to target
  3. •Create PROJECT.md with clone context
  4. •Create REQUIREMENTS.md from behaviors
  5. •Generate ROADMAP.md with eri-roadmapper

See reference.md for details.

Phase 3: IMPLEMENT - Build From Behaviors

For each module phase:

  1. •/coder:plan-phase N - Plan from BEHAVIOR.md (NOT source code)
  2. •/coder:execute-phase N - Implement in target language idioms
  3. •/coder:verify-work N - Quick verification

Key principle: Code should look like it was written by someone who never saw the source.

See reference.md for details.

Phase 4: VERIFY - Behavior Diff

For each module:

  1. •Run behavior verification
  2. •Check interface, state machine, tests, resources, ownership
  3. •Fix any ❌ violations before proceeding
  4. •Document ⚠️ manual checks

See reference.md for details.

Phase 5: COMPLETE - Finalize

  1. •Generate CLONE-VERIFICATION.md report
  2. •Tag release with /coder:complete-milestone v1.0
  3. •Final commit

See reference.md for details.

Agent Strategy

PhaseAgentModel
SCANeri-codebase-mapper, eri-behavior-extractorsonnet
PLANeri-roadmappersonnet
IMPLEMENTeri-planner, eri-executoropus (complex), sonnet (simple)
VERIFYBuilt-in verificationsonnet

Error Recovery

Check progress:

bash
python3 -m erirpg.commands.clone_behavior progress --json

Resume from failed step (not beginning).

Completion

Use templates/completion-box.md format.

Update STATE.md with:

  • •Modules cloned count
  • •Behaviors verified count
  • •Next step: Run and compare with source