The Agent Skills Directory

[SAFE]: The skill is designed to be a non-destructive auditing tool. It strictly adheres to a read-only policy for the product codebase, with the sole exception of writing its own report to a designated directory (docs/audits/).
[SAFE]: The skill explicitly forbids automatic modifications, such as fixing code, opening issues, or editing roadmaps, ensuring all actions remain under user control.
[SAFE]: It employs a transparent evidence-provenance gate, requiring cited evidence from authoritative sources (forge CLI, command outputs, repository inventories) and clearly marking unverified data.
[SAFE]: The skill uses established workflow conventions and composition of internal review packs, which are standard for high-quality agentic workflows.
[SAFE]: The disable-model-invocation: true setting in the frontmatter is a platform-level restriction that limits the agent's autonomy, further reducing the risk of unauthorized tool usage.

product-audit