The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to ingest and distill knowledge from external sources, which creates a surface for indirect prompt injection.
Ingestion points: The scripts/okb_snapshot.py script fetches content directly from arXiv URLs or reads from local files provided via the --from-file argument.
Boundary markers: The skill uses YAML frontmatter delimiters (---) to separate metadata from the raw ingested content in the 'bronze' layer files.
Capability inventory: The skill possesses the ability to read and write to the file system and perform network requests to retrieve external data.
Sanitization: There is no evidence of explicit sanitization, filtering, or 'ignore instruction' warnings applied to the external content before it is processed by the agent during the 'distill' or 'fact-check' phases.
[COMMAND_EXECUTION]: The test suite (tests/test_okb_scripts.py) utilizes subprocess.run to execute the skill's Python scripts for functional validation. While this is standard practice for testing, it involves the execution of shell commands and process spawning.
[EXTERNAL_DOWNLOADS]: The skill communicates with the official arXiv API (export.arxiv.org) to fetch research paper metadata and abstracts. This is an expected and documented behavior for the skill's research-focused use case.

okb