The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests genomic metadata and research data from external scientific APIs, creating a surface area where instructions could theoretically be embedded in the processed content.
Ingestion points: The agent processes JSON output files generated by the screen_api.py and encode_portal_api.py scripts (e.g., /tmp/search.json).
Boundary markers: There are no explicit instructions or delimiters configured to separate user-provided rules from the data fetched from the external APIs.
Capability inventory: The skill allows for local script execution via uv run and file system writes within the workspace.
Sanitization: The data is structured via JSON parsing, but textual fields like gene descriptions are not sanitized for potential natural language instructions.
[EXTERNAL_DOWNLOADS]: The skill communicates with well-known scientific repositories to retrieve genomic information.
Source URLs: It makes requests to the ENCODE Project Portal (www.encodeproject.org) and the SCREEN GraphQL API (factorbook.api.wenglab.org).
Purpose: These connections are essential for querying candidate cis-regulatory elements (cCREs) and associated biochemical signatures.
[COMMAND_EXECUTION]: The skill utilizes the uv package manager to execute its internal helper scripts.
Mechanism: Instructions recommend running uv run scripts/screen_api.py and uv run scripts/encode_portal_api.py to interact with the databases. These scripts provide a controlled interface and automate required rate-limiting (10 QPS).

encode-ccres-database