The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, including source code, build artifacts, and third-party documentation to identify security vulnerabilities.
Ingestion points: Processes repository files via Read, Glob, and Grep, and fetches external documentation using WebFetch as specified in SKILL.md and references/official-docs.md.
Boundary markers: The skill includes explicit instructions to "Quarantine and validate offline" untrusted source data and contains a section on "Approval Boundaries" to prevent unauthorized actions.
Capability inventory: Uses Write and Edit tools to generate reports and propose remediations based on the analyzed data.
Sanitization: Instructions mandate the redaction of reports and explicitly forbid the logging or inspection of actual secret values.

lucidchart-security-basics