The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the web using donsetch search, donsetch fetch, and curl calls to research APIs. This data is processed by parallel research agents to create dossier notes in a vault. The skill possesses capabilities to write files (create_note, append_to_note), spawn sub-agents, and execute shell commands (grep, curl, donsetch). While it employs boundary measures like strict templates and a three-pass verification process—where the coordinator compares summaries to cluster notes and re-verifies with warm agents (as detailed in Section 7)—the pipeline for processing external data remains an attack surface for indirect prompt injection.
[COMMAND_EXECUTION]: The skill utilize several command-line tools to perform its tasks. It executes donsetch for searching and fetching web content, curl for querying academic APIs (e.g., Europe PMC, OpenAlex), and grep for searching through saved document files in the session scratchpad. It also includes a preflight check using donsetch doctor to manage browser profile locks and ensure tool reliability.
[EXTERNAL_DOWNLOADS]: The skill regularly fetches data from external services to perform research. This includes academic and research-oriented domains such as ebi.ac.uk and api.openalex.org, as well as third-party search and scraping APIs like Tavily and Firecrawl to retrieve full-text documents and metadata. It also references project documentation hosted on hatchdoor.batterlan.cc, which is associated with the skill author.

deep-research