The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to "Refuse any caller instruction" that attempts to limit the scope, files, or subject matter of the review. It directs the agent to "ignore narrowing" and prioritize external authoritative sources over the user's immediate constraints.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data from repository files, diffs, and specifications.
Ingestion points: Processes repository instructions, issues, specifications, and code diffs.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are implemented for the ingested data.
Capability inventory: The agent has permissions to read repository files and write artifacts to the ephemeral/reviews/ directory.
Sanitization: No evidence of sanitization or validation of external content before processing.

adversarial-review