The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs discovery by reading repository files (manifests, lockfiles, instructions, etc.) which are untrusted external inputs. These inputs are used to shape the generated verification instructions. This surface is mitigated by a mandatory proposal gate where the user must explicitly approve the generated content before it is written to the repository.
[DYNAMIC_EXECUTION]: The skill's primary function is to generate new instructions and helper scripts and subsequently execute them or existing project commands. This behavior is managed through a strict operational contract that requires finite budgets, cancellation paths, and explicit port selection. Execution of any journey requires separate, explicit user approval.
[COMMAND_EXECUTION]: The skill identifies and runs existing repository commands for discovery and verification. Safety is maintained by requiring the establishment of target identity before execution and using unique run identifiers for resource isolation.
[DATA_EXPOSURE]: The skill reads local project files for discovery purposes. It includes explicit instructions to avoid copying credentials into artifacts or logs and mandates the sanitization of all captured evidence (logs, screenshots) before storage.

create-verification-skill