The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including source code diffs, pull request descriptions, and design documents. This surface could be targeted by malicious instructions embedded in the code comments or metadata being reviewed.
Ingestion points: Untrusted content enters the agent context through the file system and version control metadata as described in SKILL.md.
Boundary markers: The skill uses a highly structured report template (references/report-template.md) and a Review Coverage Ledger to maintain clear boundaries for analysis.
Capability inventory: The skill possesses the ability to read the file system, write report files, and execute a local Python validation script.
Sanitization: The skill mitigates risks by requiring findings to be grounded in authoritative product intent and verified by a coordinator agent, rather than relying solely on the input data.
[COMMAND_EXECUTION]: The skill executes local Python scripts to ensure the generated reports follow the required contract.
Evidence: SKILL.md instructs the agent to execute scripts/validate_review_report.py. The regression test scripts/test_validate_review_report.py also uses subprocess.run to verify the validator's logic.
Security Assessment: These commands are scoped to the skill's own internal validation logic. The execution is performed safely using sys.executable and avoids shell invocation, preventing command injection vulnerabilities.

code-review