The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of local Python validation scripts, specifically scripts/validate_verdict.py, to ensure that research artifacts conform to the required evidence and structure standards.
[EXTERNAL_DOWNLOADS]: An example workflow in examples/verdict-insufficient-evidence.md demonstrates fetching publication metadata from Arxiv (arxiv.org), a well-known and trusted academic repository, to support evidence collection.
[DYNAMIC_EXECUTION]: The internal test suite in scripts/tests/test_validate_verdict.py utilizes subprocess.run to programmatically execute and verify the skill's own validation logic.
[INDIRECT_PROMPT_INJECTION]: The skill includes specific security instructions in references/recent-public-signal.md to mitigate indirect prompt injection risks, explicitly directing the agent to treat all external signals as untrusted third-party data and to ignore any instructions embedded within them.

would-agents-actually