The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from documentation files and uses that information to influence its research behavior and file editing tasks.
Ingestion points: AGENTS.md files located within the project repository are read to extract context and conventions.
Boundary markers: The skill instructions do not include markers to delimit external content or warnings to ignore potential instructions embedded within the documentation files.
Capability inventory: The agent has access to Read, Glob, and Edit tools, enabling it to modify repository files based on the context it retrieves.
Sanitization: There is no evidence of sanitization or validation logic to ensure that the content read from documentation does not contain malicious instructions.

research-compound