The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local helper scripts, such as fetch_sources.py and report_lint.py, using python3 or uv run. These scripts are part of the skill's assets and perform specialized tasks like querying scholarly APIs and validating results.\n- [EXTERNAL_DOWNLOADS]: The skill fetches metadata and abstracts from well-known scholarly services including arXiv, PubMed, Crossref, OpenAlex, and Semantic Scholar. These connections are strictly limited to technical data retrieval for the purpose of grounding the audit's findings in peer-reviewed literature and technical standards.\n- [INDIRECT_PROMPT_INJECTION]: The skill is exposed to untrusted data from audited repositories and external scholarly sources. It implements significant architectural safeguards against manipulation, such as a deterministic evidence logging system that cross-references all claims against actual network logs, and a multi-agent adversarial refutation pass that subjects every finding to independent verification before reporting.\n- [DYNAMIC_EXECUTION]: The skill's evaluation suite uses dynamic script loading and evaluation to verify the correctness of the report linter and workflow aggregation logic. These patterns are limited to the testing and evaluation components of the repository.

base-in-reality