Agent Skill Bridge Review

Pattern declaration

Primary pattern: Reviewer Secondary patterns: Inversion Why: ブリッジモジュールのパス安全性・往復忠実性・バリデーション正当性をチェックリスト型で評価するが、対象ファイル外の変更では実行不要

Guidance

• Path traversal: assertSafePath must reject ids containing .., /, or NUL bytes. Any new file-write path must route through this function or sanitizeSkillId.
• Round-trip fidelity: export followed by re-import must preserve name, description, and body content. Verify that new fields added to conversion logic do not break this property.
• Auto-fill correctness: convertAgentSkillToRR must not overwrite explicitly provided values (id, category, phase, applyTo). Defaults apply only when the field is absent.
• Validation gate: both strict and loose validation paths must be exercised. The errors array must capture invalid skills without aborting the entire import batch.
• Test coverage: every newly exported function from the module must have at least one node:test case.

Non-goals

• Do not critique the RR skill schema itself; focus on bridge logic only.
• Do not flag auto-fill behaviour as missing validation when the converted metadata passes the strict RR schema.

Pre-execution Gate / 実行前ゲート

このスキルは以下の条件がすべて満たされない限りNO_REVIEWを返す。

• 差分にAgent Skillブリッジ関連ファイル（agent-skill-bridge.mjs, validate-agent-skills.mjs, agent-skill-loose.schema.json等）の変更が含まれている
• 差分がコメント・JSDoc・空白のみの変更ではない
• inputContextにdiffが含まれている

ゲート不成立時の出力: NO_REVIEW: agent-skill-bridge — Agent Skillブリッジ関連の実質的なコード変更が検出されない

False-positive guards

• If test fixtures are added without corresponding logic changes, that is expected (test-only commits).