The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill is designed to read AGENTS.md files from the local environment to extract context and patterns. This creates a surface for indirect prompt injection where an attacker could place malicious instructions inside a repository's AGENTS.md file.
Ingestion points: Reads AGENTS.md files during the 'Pre-Research' phase using the Read tool.
Boundary markers: None. The instructions do not specify any delimiters or warnings to ignore instructions found within the files being read.
Capability inventory: Uses Edit and Glob tools to modify or create files (AGENTS.md) and search the directory structure.
Sanitization: None. Extracted content is processed and potentially appended to other files without validation or escaping.
Command Execution (LOW): The skill documentation includes bash examples for finding files (e.g., find . -name "AGENTS.md"). While these are provided as examples for the agent's logic, they indicate an expectation of shell access to perform file operations.

research-compound