The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the parallel-cli tool using the Python subprocess module to perform search and extraction tasks.
Evidence: Found in scripts/research_lookup.py. The implementation uses argument lists rather than shell strings, which is a safe practice that prevents shell injection.
[EXTERNAL_DOWNLOADS]: The skill interacts with external APIs to retrieve scholarly data and research evidence.
Evidence: Connects to api.parallel.ai and openrouter.ai (for Perplexity) as documented in SKILL.md and implemented in scripts/research_lookup.py. These are well-known services relevant to the skill's purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes external research data (titles, abstracts, and excerpts) which could potentially contain malicious instructions intended to influence the agent's behavior.
Ingestion points: External data is ingested through search and extraction results in scripts/research_lookup.py.
Boundary markers: SKILL.md contains an explicit warning to treat all returned web content as untrusted data rather than instructions.
Capability inventory: The skill can execute local CLI tools via subprocess and write research artifacts to the local file system.
Sanitization: Employs structured metadata mapping and normalization in scripts/manuscript_packet.py to structure the untrusted content.

research-lookup