The Agent Skills Directory

[SAFE]: The skill implements strong procedural guardrails to ensure its operations are targeted and verifiable. It requires users to confirm a specific scan scope before proceeding, preventing blind repository-wide scans that could lead to excessive resource consumption or noise.
[SAFE]: The workflow separates discovery from remediation, ensuring the agent only identifies issues without making unauthorized changes to the codebase. Findings are recorded in a standardized, local directory structure (.codestable/audits/).
[DATA_EXFILTRATION]: No network-enabled tools or external exfiltration patterns were identified. The skill focuses entirely on reading local code and writing markdown reports.
[PROMPT_INJECTION]: The instructions do not contain any patterns attempting to override safety filters, extract system prompts, or bypass agent constraints. It includes specific 'Guardian Rules' to maintain intended behavior.
[COMMAND_EXECUTION]: While the skill scans for security risks like command injection in user code, the skill itself does not execute arbitrary shell commands or untrusted scripts. It relies on standard tools like Grep and Read for analysis.

cs-audit