The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as web pages, search snippets, and API payloads to verify claims. It proactively mitigates the risk of indirect prompt injection through a multi-layered defense strategy.
Ingestion points: External content is fetched via search and browsing tools as defined in SKILL.md and references/evidence-policy.md.
Boundary markers: The skill includes an explicit 'Untrusted Content Policy' in references/evidence-policy.md which instructs the agent to never follow instructions found inside fetched content and to treat them strictly as data.
Capability inventory: The skill is primarily read-only, focusing on information retrieval and response validation. It does not perform high-impact actions like system modifications or financial transactions.
Sanitization: It employs a 'Revision Screening Gate' in references/repair-and-integration.md to ensure suggested edits are grounded in evidence and do not introduce unverified content or follow embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill references several external URLs in references/external-sources.md for methodology guidance. These targets include well-known educational institutions (Purdue University, CSU Chico), established security organizations (OWASP), and reputable technology/information literacy sites (Simon Willison, Hapgood). These references are used for static background guidance on information evaluation and do not involve the execution of remote code.

recency-guard