The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The agent is configured to read the full project codebase, which may contain untrusted third-party content. It also has access to the bash tool and file-writing capabilities in agents/openai.yaml. The absence of explicit boundary markers or instructions to ignore embedded instructions in the system prompt creates a potential surface for indirect prompt injection.
Ingestion points: Full project read access via read_file, grep, and codebase_search.
Boundary markers: None defined in the system prompt.
Capability inventory: Shell execution via bash, npm, python, pytest, jest, and cargo.
Sanitization: No input sanitization or filtering logic is specified.
[COMMAND_EXECUTION]: The skill requests access to powerful shell commands including npm, python, cargo, and bash. While these are intended for development and test execution, they provide broad execution capabilities within the local environment.

test-gen