The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process adversarial security content from external sources, such as exploit write-ups and vulnerability reports. This represents a potential surface for indirect prompt injection. The skill addresses this by mandating the use of a content-sanitization guardrail and providing explicit instructions to treat all ingested content as data rather than instructions, prohibiting the execution of code or the following of links found within the content.
[COMMAND_EXECUTION]: The test script scripts/test_validate_security_prior_art.py uses subprocess.run to invoke the local validator script during automated testing. This execution is restricted to the local Python interpreter and the provided script using controlled argument lists, which is a standard and safe testing practice.
[EXTERNAL_DOWNLOADS]: The skill performs downloads from well-known and trusted security corpora and vendor advisory feeds, including MITRE (CWE/CAPEC), FIRST (EPSS), and official advisory channels for major cloud platforms like AWS, Google Cloud, and Microsoft. These downloads are central to the skill's research function and are required to pass through a sanitization process.

security-prior-art-survey