The Agent Skills Directory

[COMMAND_EXECUTION]: The skill' s testing scripts (scripts/test_prove_ac.py and scripts/mutate_prove_ac.py) utilize subprocess.run() to execute other local Python scripts included in the skill. This is used exclusively for internal unit testing and verification of the skill's logic.
Evidence includes calls to sys.executable with local paths for prove_ac.py, scan_bypasses.py, and review_gate.py.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface as it reads and parses repository files (such as requirements.md, spec-coverage.md, and code files) to find acceptance criteria tokens.
Ingestion points: scripts/prove_ac.py uses Path.read_text() and os.walk() to ingest repository content.
Boundary markers: The skill requires findings to include internal evidence references (path:line) to ensure the agent's claims are grounded in the processed data.
Capability inventory: The skill is primarily read-only on product files, with write access limited to a local findings.json results file. It has no network capabilities.
Sanitization: Input data is processed via strict regular expressions for token extraction rather than being directly interpreted as commands.
[EXTERNAL_DOWNLOADS]: SKILL.md contains a link to an external GitHub repository (github.com/tech-leads-club/agent-skills) as a documentation reference for its implementation logic. The instructions explicitly forbid downloading or pasting content from this URL, treating it solely as a conceptual reference.
[SAFE]: The skill's scripts follow security best practices by skipping sensitive directories such as .git, node_modules, and .venv during file system crawls. No hardcoded credentials or obfuscation techniques were identified.

ns-judge