The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell script evals/run-static-checks.sh used for testing and validation. This script executes bash and python3 to run a static analyzer (check-skill-static.py) located in the repository's root. This is a standard development practice for ensuring skill integrity.
[COMMAND_EXECUTION]: In Step 9 of the activation workflow, the skill is instructed to write findings and workflow state to the local filesystem (e.g., docs/audits/ or audit-artifacts/). This is part of the core functionality to track and resume audit states.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as defined in Category 8, because its primary function is to ingest and analyze potentially untrusted technical artifacts provided by users (such as CLI help text, API response bodies, or code snippets).
Ingestion points: Documentation files, CLI help output, and API response contracts loaded during the audit process.
Boundary markers: No explicit instructions were found to use XML tags or other delimiters to isolate untrusted data during analysis.
Capability inventory: The skill has capabilities for file system writes and local command execution (via eval scripts).
Sanitization: No specific sanitization or filtering of the processed developer artifacts is mentioned in the instructions.

dx-audit