The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions designed to manipulate the agent's release and deployment decisions.
Ingestion points: The workflow in SKILL.md ingests repository state, exact diffs, review findings, and CI test results.
Boundary markers: The instructions lack explicit boundary markers or directives to the model to ignore instructions embedded within the diffs or comments it analyzes.
Capability inventory: The skill is authorized to perform Git operations (commit, push, merge) and deployment actions through CI/hosting connectors.
Sanitization: No explicit sanitization or filtering of external content is defined in the instructions.

starlight-ship