The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands such as grep, glob, and a project-specific Python script .claude/scripts/code_graph. These are standard tools used for code analysis within the project environment and do not involve untrusted input in shell execution contexts.
[PROMPT_INJECTION]: The skill uses strong instructional language (e.g., "MANDATORY", "NON-NEGOTIABLE", "IMPORTANT") to enforce strict debugging standards, human-in-the-loop checkpoints, and planning steps. These instructions reinforce safety and quality protocols rather than attempting to bypass model constraints or safety filters.
[DATA_EXFILTRATION]: Analysis is restricted to local codebase files and project documentation. The skill writes its findings and reports to local directories like plans/reports/ and .ai/workspace/analysis/. There are no network operations targeting external domains or untrusted servers.
[EXTERNAL_DOWNLOADS]: The skill does not perform any remote package installations or download external scripts. It relies on pre-existing local scripts and standard system utilities.
[SAFE]: The skill includes robust safety features, such as mandatory plan validation, red flag stop conditions that escalate to the user via AskUserQuestion, and a requirement for post-fix verification.

fix