Gemini Skill Editor

Gemini Skill Editor

Applies a structured improvement plan to Gemini skill files. It is deliberately narrower than arbitrary skill editing: it consumes plans from gemini-skill-improvement-planner and updates the named target skills.

Core Rules

• Read the improvement plan and target SKILL.md before editing.
• Use the active session's file-editing tool for manual edits.
• Edit only skills named by the plan.
• Default target set is gemini-* skills. Do not edit the original Claude-oriented skills unless the plan explicitly names them and the user confirms that scope.
• Preserve skill frontmatter validity.
• Do not push or commit unless the user explicitly requests it.
• Archive consumed reflections only after all recommendations citing them succeeded.

Inputs

• Plan path: explicit path, or latest ~/.gemini/skills/gemini-skill-improvement-planner/plans/plan-v*.md.
• --dry-run: parse and report intended edits without changing files.

If no plan path is explicit, first check the current repo and branch handoff from gemini-skill-improvement-planner using gemini-config/shared/runtime-state.md: read ~/.gemini/skills/gemini-skill-improvement-planner/handoffs/<repo_hash>/<branch_slug>/latest.md, validate from, repo, repo_root, branch, branch_slug, commit, and artifact, then use the artifact only if it exists under the current repo root. Ignore missing or mismatched handoffs unless the user explicitly asks to reuse cross-branch state.

Workflow

1. Resolve and read the plan.
2. Parse:
   • reflections_consumed;
   • recommendations by skill;
   • cross-cutting recommendations;
   • contradictions.
3. If contradictions exist, stop and ask the user how to resolve them unless the plan already contains a resolution.
4. Validate target skills:
   • source path under gemini-config/skills/<skill>/SKILL.md when working in this dotfiles repo;
   • symlink/runtime path under ~/.gemini/skills/<skill>/SKILL.md only when no source path exists.
5. For --dry-run, report the target files and recommendation summaries, then stop.
6. Apply recommendations:
   • group changes per target skill to avoid conflicting edits;
   • keep SKILL.md concise;
   • move lengthy examples into references/;
   • update agents/openai.yaml when display metadata becomes stale.
7. Validate:
   • YAML frontmatter parses;
   • name matches the skill directory intent;
   • description clearly states trigger scope and non-scope;
   • referenced files exist.
8. Archive reflections:
   • move successfully consumed reflection files to an archive/ directory under the same repo and branch subtree;
   • leave reflections in place for failed recommendations.

Failure Policy

• Malformed plan: stop and report exact parse failure.
• Missing target skill: mark that recommendation failed; continue only if other independent targets remain.
• Patch conflict: re-read the file, adjust once, then report if still blocked.
• Validation failure: fix if local to the edit; otherwise roll forward with a clear report and do not archive affected reflections.

Closeout

Report:

• applied recommendations;
• skipped or failed recommendations;
• files changed;
• reflections archived;
• validation commands run.

If writing self-improvement state, follow gemini-config/shared/runtime-state.md and use Gemini paths only:

• Reflection: ~/.gemini/skills/gemini-skill-editor/reflections/<repo_hash>/<branch_slug>/<run_id>.md
• Handoff: ~/.gemini/skills/gemini-skill-editor/handoffs/<repo_hash>/<branch_slug>/<run_id>.md
• Latest handoff pointer: ~/.gemini/skills/gemini-skill-editor/handoffs/<repo_hash>/<branch_slug>/latest.md

Handoff frontmatter must include from: gemini-skill-editor, timestamp:, repo:, repo_root:, branch:, branch_slug:, commit:, run_id:, and artifact:. Update latest.md with the same handoff content.