The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/run.py script dynamically determines the repository root by searching up the directory tree for an AGENTS.md file and subsequently modifies sys.path to include this computed path. This technique enables the dynamic loading of the tooling package from a non-static location, which is a common but risky practice for library resolution.
[INDIRECT_PROMPT_INJECTION]: The skill audits content from potentially untrusted sources, creating an injection surface.
Ingestion points: The script reads untrusted content from output/DRAFT.md and related outline files in scripts/run.py.
Boundary markers: The script does not utilize boundary markers or explicit instructions to ignore embedded commands within the draft text.
Capability inventory: The skill possesses file-writing capabilities to generate the output/AUDIT_REPORT.md and associated scorecard files.
Sanitization: The _examples function in scripts/run.py extracts raw snippets from the draft content and includes them in the audit report without sanitization or escaping, allowing potentially malicious instructions to persist in the output.

pipeline-auditor