The Agent Skills Directory

[SAFE]: The skill functions as a purely local diagnostic utility. It does not initiate network connections, download external scripts, or attempt to access sensitive system files such as credentials or environment variables.
[COMMAND_EXECUTION]: File modifications are handled using the str_replace tool. The instructions explicitly enforce a security-conscious workflow: the agent identifies issues, provides a 'flagged-issues report', and must wait for explicit author confirmation before performing any edits. This prevents autonomous or unauthorized modification of the manuscript.
[PROMPT_INJECTION]: The skill instructions are clear and focused on the auditing task. There are no attempts to bypass safety filters, extract system prompts, or override agent constraints.
[DATA_EXFILTRATION]: The skill only reads project-specific files (SKILL.md, references/*.md, and audit-tracker.md). No data is transmitted externally, and no sensitive information is targeted.

prose-mechanics