The Agent Skills Directory

[SAFE]: The skill performs project auditing through deterministic analysis of local repository files. No malicious network activity, credential harvesting, or unauthorized command execution was identified.\n- [SAFE]: The audit tool ingests untrusted data from the repository being analyzed (e.g., README.md, feature documentation) to calculate scores and completion rates. While this creates a surface for indirect prompt injection, the logic is limited to deterministic regex counting and line-length checks, which significantly mitigates the risk of the agent being manipulated by malicious content within the audited files.\n
Ingestion points: scripts/audit.js reads README.md and Markdown files within the docs/features/ directory of the target project.\n
Boundary markers: The script generates structured JSON output which the agent parses, providing a layer of separation between raw file content and agent instructions.\n
Capability inventory: The agent can suggest follow-up actions like /update-docs, /codex-test-gen, or /create-request based on the audit results.\n
Sanitization: The tool uses specific regex patterns (e.g., - [x], #{1,3}\s) to extract data, rather than interpolating raw user text into control logic.

project-audit