The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, such as package registries and changelogs, to evaluate dependency upgrades, which constitutes a potential ingestion surface for indirect instructions. * Ingestion points: Dependency version metadata and external changelogs retrieved during the audit process. * Boundary markers: Employs mandatory Git worktree isolation to prevent interference with the main workspace. * Capability inventory: Instructs the agent to perform dependency installations and execute test/lint suites for verification. * Sanitization: Emphasizes isolation boundaries and evidence-backed verification rather than explicit string sanitization of incoming metadata.

inspect-deps