reflect-code-skills

Imported from bew/dotfiles (opencode/skills/reflect-code-skills/SKILL.md). Install upstream with npx skills add bew/dotfiles --skill reflect-code-skills. Copyright stays with the author.

Goal

Surface patterns and decisions from the current code-writing session that are not yet encoded in the active coder-* skills, and ask the user whether to add them.

When to load

Load this skill when all of the following are true:

• A coder-* skill was active during the session.
• A code file was written or meaningfully iterated.
• The user explicitly signals the script is done (e.g. "looks good", "done", "ship it").

Do not load speculatively or mid-iteration.

Steps

1. Scan the session — Review what was written or changed in the script. Identify anything that:

   • Is a deliberate stylistic or structural choice.
   • Deviates from or extends the active coder-* skills.
   • Was improvised without a documented rule backing it.
   • Came up as a decision point (e.g. "should I use X or Y?").

2. Identify candidate rules — For each pattern found, decide:

   • Is it general enough to belong in coder-generic?
   • Is it language-specific and belongs in the active lang skill?
   • Is it a one-off, not worth encoding?

3. Ask the user — Present candidates as a short list. For each, name: what the pattern is, which skill it would go into, and why. Ask the user to confirm which ones to capture.

4. Summarize — After the user responds, output a brief summary:

   • What will be added and where.
   • What was consciously left out and why.

   If no candidates were found, say so briefly — do not invent friction.

Rules

• Never edit skill files silently — always confirm with the user first.
• Never add rules that are already covered in coder-generic or the active lang skill.
• Keep candidates concrete: name the actual pattern observed, not an abstraction of it.
• One candidate per bullet — do not bundle unrelated patterns together.