The Agent Skills Directory

[DYNAMIC_EXECUTION]: The scripts/research_validation.py script performs dynamic loading and execution of Python code from a path retrieved from the JEV_PRESCREEN_PATH environment variable. It uses importlib.util.spec_from_file_location to load a module and subprocess.run to execute a script with the current Python interpreter. This creates a risk of arbitrary code execution if the environment variable points to a malicious or unverified file.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and synthesize information from the web, which introduces a surface for indirect prompt injection.
Ingestion points: Web page text extracted using web_extract during Step 3b of the research loop in SKILL.md.
Boundary markers: None; the instructions do not establish clear boundaries or specific directives for the agent to ignore potential instructions embedded in the source data.
Capability inventory: The agent can perform web searches, write files, and execute local scripts (specifically scripts/research_validation.py).
Sanitization: The skill filters content for relevance and quality but does not sanitize content against adversarial prompts or instructions.
[COMMAND_EXECUTION]: The scripts/research_validation.py script uses the subprocess module to run external commands. While currently intended to run an advisory pre-screen script with the Python interpreter, the implementation allows the execution of arbitrary scripts provided via environment variables, posing a risk of command execution abuse.

deep-research