The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow involves processing untrusted data from multiple sources which could contain malicious instructions designed to subvert the agent behavior. * Ingestion points: Data enters the context via git diff outputs, pull request descriptions, commit messages, and external specification URLs supplied by users or found in repository metadata. * Boundary markers: The instructions do not mandate the use of delimiters or provide warnings to ignore instructions that might be embedded in the code or external specifications being reviewed. * Capability inventory: The skill utilizes file system access to read repository content and network access to fetch external documentation and specifications. * Sanitization: There is no requirement to sanitize, escape, or validate the content of the external specifications or pull request discussions before they are interpreted by the agent.

code-reviewing