The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where an agent may ingest untrusted datasets and execute recorded commands via datalad rerun.
Ingestion points: datalad clone from arbitrary URLs in SKILL.md and references/data-access.md.
Boundary markers: The skill mentions tool-specific delimiters (=== Do not change lines below ===) used in commit messages to store execution records.
Capability inventory: The skill has access to the Bash tool and documents commands like datalad run and containers-run which execute arbitrary shell and containerized commands.
Sanitization: No sanitization is mentioned for command strings extracted from dataset provenance records.
[COMMAND_EXECUTION]: The skill uses the Bash tool to perform operations and documents the use of datalad run, rerun, and containers-run for executing shell commands and containerized software environments.
[EXTERNAL_DOWNLOADS]: The skill downloads software packages from PyPI (git-annex, datalad) and fetches scientific datasets from services such as GitHub, OpenNeuro, and DANDI. It also retrieves container images from Docker Hub and Singularity Hub.

datalad