The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches a CSV catalogue from https://depmap.org/portal/api/no-captcha/download/files. This is a well-known scientific service hosted by the Broad Institute, and the operation is central to the skill's stated purpose.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the DepMap portal, creating an indirect prompt injection surface.\n
Ingestion points: Data is fetched in scripts/depmap_data.py via the fetch_catalogue function.\n
Boundary markers: The skill checks Content-Type and content starts to ensure the response is not a redirected HTML verification page.\n
Capability inventory: The skill's capabilities are limited to data analysis, local file reads, and network metadata retrieval. It lacks shell execution or file-writing permissions.\n
Sanitization: The code performs rigorous validation, including MD5 checksum verification for downloaded files, regex-based validation of Model IDs (ACH-...) and gene labels, and numeric type enforcement for matrices.\n- [COMMAND_EXECUTION]: No instances of shell command execution or arbitrary code execution were found. All logic is implemented using standard Python scientific libraries.\n- [CREDENTIALS_UNSAFE]: No hardcoded secrets, API keys, or private tokens are present. The skill documentation explicitly guides users to use the portal for verification rather than attempting to bypass security mechanisms.

depmap