The Agent Skills Directory

[SAFE]: The skill defines a procedural workflow for implementing corrections found during a project audit. It provides structured templates for updating project artifacts such as CLAUDE.md, SKILL.md sections, and changelogs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests audit-report.md as a functional specification. While reading external files constitutes an indirect prompt injection surface, the skill guides the agent using specific templates and idempotency markers (e.g., HTML comments) to ensure actions remain within the intended scope of project maintenance.
[SAFE]: The skill includes an explicit safety policy regarding its execution environment, specifically forbidding automated writes to the global skills directory (~/.claude/skills/). This mitigates risks related to unauthorized persistence or environment modification.
[SAFE]: The documentation mentions an assembly pattern using an external script (install.sh) to build the skill logic. This appears to be a standard build-time process for the development framework and does not involve runtime remote code execution or untrusted dependency fetching.

project-fix