The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-generated or agent-generated content from the workspace to perform quality checks.
Ingestion points: The scripts/run.py script reads multiple files from the output/ and output/trace/ directories, including REPORT.md, REVIEW.md, and various trace tables.
Boundary markers: The script does not utilize explicit delimiters to isolate the content of the deliverables during evaluation.
Capability inventory: The script is restricted to local file system read/write operations and internal module imports.
Sanitization: Content validation is performed using basic string matching and regular expressions to identify required headers and block templated phrases.
[DYNAMIC_EXECUTION]: The script dynamically adjusts the Python search path (sys.path) to locate and import internal tooling.
The script identifies the repository root by traversing the file system upward from its own location to find an AGENTS.md file. This is a standard utility pattern for resolving internal dependencies within a project repository.

deliverable-selfloop