The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses git diff origin/main...HEAD to identify modified files and retrieve the code changes necessary for the architectural review pass.
[PERSISTENCE]: Analysis results and diff hashes are saved to .claude/state/code-hygiene/<branch-slug>.json. This local state management is used to implement a caching mechanism that skips redundant reviews for unchanged code patches.
[INDIRECT_PROMPT_INJECTION]: This skill possesses an attack surface for indirect prompt injection as it ingests and processes untrusted source code from the repository.
Ingestion points: The skill reads code content and documentation (Docusaurus pages, PRDs) through git diffs and file system access.
Boundary markers: There are no explicit delimiters or instruction-ignore warnings defined for the ingested code content.
Capability inventory: The skill is primarily restricted to reporting findings via the ReportFindings tool. It has a limited capability to mechanically wrap errors with static sentinels, which is a localized code modification.
Sanitization: No specific sanitization or escaping of the input code is performed prior to the agent's architectural review pass.

code-hygiene