The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and diffs as its primary input. Ingestion points: The skill reads 'caller's files or diff' or 'current diff against the base branch' to identify comments for removal. Boundary markers: There are no explicit instructions to the agent to treat input text as non-executable data, creating a surface for indirect prompt injection. Capability inventory: The skill has capabilities to modify files, delegate tasks to other agents like 'comment-sicko', and generate new code ('encode then delete') based on the content of comments. Sanitization: The skill relies on interactive user approval (Step 5) as a mitigation before committing changes based on constraint comments, but lacks automated sanitization of the input text itself.

no-comments