The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external content, including pull request diffs, metadata, and comments. This data could potentially contain malicious instructions intended to manipulate the agent's review verdict or subsequent actions. * Ingestion points: The scripts/collect_context.sh script fetches PR metadata, diff patches, and issue/review comments using the GitHub CLI (gh). * Capability inventory: The skill has access to the Bash tool and executes local Python and shell scripts to process data. It also dispatches sub-agents for specialized review tasks. * Boundary markers: The prompt templates in references/external_prompt_template.md use structured delimiters like , <grounding_rules>, and , which help separate instructions from data but do not fully eliminate the risk of adversarial content in the interpolated source. * Sanitization: There is no evidence of explicit sanitization or filtering of the external data before it is presented to the agent for analysis.
[COMMAND_EXECUTION]: The skill legitimately uses system commands to perform its primary function of code analysis and repository interaction. * Evidence: scripts/collect_context.sh and scripts/review_scope.py invoke git and gh commands via subprocesses to retrieve repository history, diffs, and PR details.

full-review