The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to execute local commands required for its operation. Specifically, scripts/evidence.py calls git show to verify code findings, and scripts/pool.py spawns claude -p processes to perform automated judgment of review items. These are controlled executions of standard tools intended for the skill's primary functionality.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data that is interpolated into prompts for the LLM judge, creating a surface for indirect injection.
Ingestion points: Data is read from findings.jsonl, issues.jsonl, and source code files from the target git repository.
Boundary markers: The judge prompt in prompts/judge.txt uses template placeholders (e.g., $findings, $summary) and explicit headers to separate instructions from untrusted content.
Capability inventory: The skill possesses the ability to execute git and claude commands via subprocess and write JSON/Markdown ticket files to the local file system.
Sanitization: The pool.py script enforces a strict JSON schema on model responses to prevent malformed or malicious outputs. Additionally, wiki.py provides character escaping when converting text to Jira wiki markup.

review-tickets