The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local bash scripts bundled with the skill (scan.sh, quick-diff.sh, save-results.sh) to perform inventory and manage results. The scripts use standard Unix utilities (jq, awk, find) and follow secure coding practices, such as using jq with explicit argument passing to prevent shell injection when handling file paths and metadata.
[INDIRECT_PROMPT_INJECTION]: The skill audits other skill files, which involves processing untrusted data that could contain malicious instructions designed to influence the evaluation subagent. The skill mitigates this by using structured prompts with clear boundary markers ([INVENTORY], [CHECKLIST]) and strictly requiring human confirmation before executing any recommended actions like archiving or deleting files.

skill-stocktake