The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill executes a shell command automatically when the SKILL.md file is loaded by the agent. This is used to generate a dynamic Table of Contents.
Evidence: The SKILL.md file contains a command block starting with ! that attempts to locate and execute skill-toc from the plugin-utils directory or the user's PATH.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external legal data, creating a potential surface for indirect prompt injection if the ingested text contains malicious instructions.
Ingestion points: scripts/bulk_fetch.py downloads data from remote S3 buckets using urllib.request.
Boundary markers: Absent. No specific delimiters or warnings are used to signal the agent to ignore instructions embedded in the legal text.
Capability inventory: The skill script possesses file write capabilities (open(a.out, "w")) and network read access.
Sanitization: No sanitization or filtering of the dataset content is performed beyond standard CSV parsing.
[EXTERNAL_DOWNLOADS]: The skill fetches bulk data files from an external storage service.
Evidence: scripts/bulk_fetch.py connects to https://com-courtlistener-storage.s3-us-west-2.amazonaws.com to download legal research snapshots. This is the established official storage for Free Law Project's CourtListener data.

courtlistener