The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes session logs and repository files (such as CLAUDE.md, AGENTS.md, and CI workflows) to identify improvement candidates. These sources represent an ingestion surface where data could potentially contain instructions intended to influence the agent's retrospective analysis.
Ingestion points: Session logs on the local machine and various repository configuration or documentation files (SKILL.md).
Boundary markers: The skill lacks explicit instructions or delimiters to ignore instructions embedded within the logs or repository files.
Capability inventory: The skill primarily generates suggestions for the user. It is restricted by disable-model-invocation: true and does not include subprocess calls, network exfiltration, or file-write capabilities in its instructions.
Sanitization: No explicit sanitization of log content or repository file data is described.

retro