The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes various research-related files from a project directory, creating an attack surface for instructions embedded in data.\n
Ingestion points: Reads workflow.json, project.json, and research phase files (e.g., in Literature/, Analysis/) located in the project root.\n
Boundary markers: Lacks explicit delimiters or instructions to ignore embedded commands in processed files.\n
Capability inventory: Capable of writing project files and executing local scripts via subprocesses.\n
Sanitization: Uses a strict path validation function (inside) that blocks path traversal and prevents access to .env or .git files.\n- [COMMAND_EXECUTION]: The scripts/leader.py script executes sibling research tools using subprocess.run to restore project context.\n
Evidence: Uses sys.executable and passes arguments as a list, which prevents shell-injection attacks.\n- [DYNAMIC_EXECUTION]: The skill dynamically locates and runs internal scripts based on a registry of research modules.\n
Evidence: The resume command in leader.py calculates script paths at runtime to invoke the memory.py module.

light-leader