The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the installation of the 'parallel-cli' dependency by piping a remote script directly into the shell: curl -fsSL https://parallel.ai/install.sh | bash. This occurs in both SKILL.md and README.md. The domain 'parallel.ai' is not recognized as a trusted organization or well-known service, presenting a significant supply-chain risk.
[COMMAND_EXECUTION]: The script scripts/generate_schematic.py uses subprocess.run() to execute the auxiliary script generate_schematic_ai.py. This pattern of spawning subprocesses to handle user-provided diagram descriptions increases the risk of command injection if the input is not strictly validated.
[DATA_EXFILTRATION]: The skill transmits user research queries, diagram descriptions, and potentially sensitive scholarly data to external services including api.parallel.ai and openrouter.ai. While this behavior is documented in the skill's metadata, these external endpoints are not whitelisted and represent data exposure points for the user's research context.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core research and generation workflows. 1. Ingestion points: User-provided research queries in lookup.py and diagram descriptions in scripts/generate_schematic.py. 2. Boundary markers: Absent; user input is interpolated directly into system prompts for Parallel and Perplexity models. 3. Capability inventory: Includes network operations (requests, openai), subprocess execution, and local file writing. 4. Sanitization: Absent; the code does not perform escaping or validation of the ingested strings before they are sent to downstream LLM backends.

research-lookup