The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core function is to process third-party feedback, creating a surface for indirect prompt injection.
Ingestion points: Feedback is gathered from conversation context, local file paths, or external URLs in Step 1.
Boundary markers: The instructions lack requirements for delimiters or warnings to subagents about ignoring embedded instructions within the feedback content.
Capability inventory: The skill utilizes subagents, reads files, performs network fetches, and executes git commands.
Sanitization: No sanitization or validation of the feedback content is described before it is passed to interpretation agents.
[COMMAND_EXECUTION]: The skill uses git commands to read repository content.
Evidence: Step 2 explicitly directs the agent to use git show to inspect code at specific references.
[EXTERNAL_DOWNLOADS]: The skill can perform network operations to retrieve content from external URLs.
Evidence: Step 1 includes logic to fetch content if a URL is provided as a source for feedback.

interpret-feedback