The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's workflow requires the agent to ingest and analyze potentially untrusted external data, including code diffs, source files, and commit history. This data could contain adversarial content designed to mislead the agent during the refactoring process. * Ingestion points: The workflow in SKILL.md (Steps 1, 2, 3, and 4) instructs the agent to read branch diffs, subsystem files, and project history. * Boundary markers: There are no specific instructions or delimiters defined to isolate the untrusted code content from the agent's instructions. * Capability inventory: The skill empowers the agent to plan and apply code modifications, grep repository headers, and generate new tests. * Sanitization: The skill does not provide any instructions for sanitizing or validating the ingested code or diff content.

aiclean