The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The files references/audit-checklist.md and evals/evals.json contain hardcoded strings that match the format of live API credentials (e.g., sk_live_4eC39HqLyjWDarjtT1zdp7dc and sk_live_9fKq2pRt7LwZ3xVn8MdE4cYb). Although these are explicitly marked as negative examples in the documentation to illustrate what the agent should detect and remove, they remain hardcoded literals within the skill's source files.
[COMMAND_EXECUTION]: The skill workflow involves executing a bundled Python script, scripts/sample-audit.py, to perform a mechanical census and linting of the documentation corpus. The script uses the Python standard library to walk the file system and perform regex-based scanning of files in a user-specified directory.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted external data (a documentation directory provided by the user). This data could contain malicious instructions designed to bypass the audit rubric, influence the agent's scoring, or override its instructions. The skill does not define specific boundary markers or sanitization logic to mitigate these indirect injection attempts when the agent processes the code samples.
Ingestion points: The docs-dir provided by the user is walked by the scripts/sample-audit.py script and its content is subsequently analyzed by the agent.
Boundary markers: None identified in the skill instructions to separate audited content from the agent's operating instructions.
Capability inventory: The skill uses file-read operations (os.walk, open) and executes a bundled Python script to process external files.
Sanitization: The skill lacks explicit sanitization or escaping mechanisms for the content it reads from the audited documentation corpus.

docs-code-sample-standards