The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/research_lookup.py interacts with the Parallel platform by executing parallel-cli through subprocesses.\n
Evidence: The _run_parallel_cli method uses subprocess.run with arguments passed as a list, which prevents shell-based command injection. The CLI is called with fixed commands and user-controlled queries as arguments.\n- [INDIRECT_PROMPT_INJECTION]: The skill fetches scientific abstracts and study excerpts from external websites, creating a surface where malicious instructions could be embedded in the retrieved content.\n
Ingestion points: Web search results and batched extractions retrieved in scripts/research_lookup.py.\n
Boundary markers: The SKILL.md contains clear instructions for the agent to treat all web content as untrusted data rather than instructions.\n
Capability inventory: The skill possesses the ability to execute CLI commands, write multiple files to the disk (save_packet), and perform network operations via requests.\n
Sanitization: While the skill normalizes identifiers (DOIs, PMIDs) and study metadata, the qualitative findings are incorporated from external sources into the final Markdown/JSON reports.\n- [EXTERNAL_DOWNLOADS]: The skill's functionality relies on an external command-line interface provided by the vendor.\n
Evidence: The setup instructions in README.md and SKILL.md require installing parallel-web-tools[cli]==0.7.1.\n- [DATA_EXFILTRATION]: Research queries and manuscript context are sent to external APIs for processing.\n
Evidence: The skill communicates with api.parallel.ai for primary research and openrouter.ai as an optional fallback, which is consistent with its stated purpose.

research-lookup