The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/detect_hazards.py uses subprocess.run to execute the git binary. This is used to determine which files have changed and to retrieve diffs for analysis. The implementation is safe as it avoids the use of shell=True and passes arguments as a list.
[INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and auditing untrusted source code. This creates an attack surface where malicious instructions embedded within the analyzed code (such as in comments or string literals) could potentially influence the agent's behavior or decision-making during the review process.
Ingestion points: The scripts/detect_hazards.py script reads local source files, and the agent is instructed to perform manual code review on the same content.
Boundary markers: There are no explicit delimiters or warning instructions used to isolate the audited code from the agent's main instruction context.
Capability inventory: The skill environment has access to read files, execute the git binary, and provide text-based feedback to the agent.
Sanitization: The scanner detects specific patterns but does not escape or sanitize the content of the files before it is processed by the agent.

poka-yoke