Proposed Skills

Annotated skill definitions for Claude Code — each one a ratchet turn that closes a computation channel.

These proposals show what superpowers/skills look like when designed from ratchet-detect data. Each skill is a crystallized pattern: a computation channel that appeared frequently in conversation logs, got characterized, and now has a structured replacement. The skill definitions are shown in code blocks with annotations explaining the design reasoning.

None of these are implemented yet as Claude Code plugins. They’re proposals — the Stage 1 observation that precedes Stage 2 crystallization. Read them as annotated blueprints, not as finished tools.

• Proposed Skill: ratchet-review
• Proposed Skill: git-workflow
• Proposed Skill: build-query
• Proposed Skill: codebase-explore

The pattern

Each skill follows the same structure:

1. Identify the computation channel from ratchet-detect data (which Bash patterns dominate?)

2. Map to structured alternatives that already exist (JetSam, blq, Fledgling, built-in tools)

3. Define the routing rules (when to use the structured tool, when to fall back to Bash)

4. Annotate the reasoning (which ratchet concept drives each design decision)

The skills are themselves ratchet artifacts — the discovery phase (ratchet-detect data) crystallized into teaching artifacts (skill definitions) that route future agent behavior through data channels instead of computation channels.