The Agent Skills Directory

[SAFE]: The skill defines a testing procedure that prioritizes security and data integrity. It mandates that testing be performed in non-production environments and explicitly forbids the use of production data or the execution of real-world transactions (payments, emails) without separate authority. The instructions also enforce a read-only constraint on source code and prevent the skill from autonomously modifying code or installing additional software stacks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of branch names, pull requests, and commit hashes to resolve the testing candidate. While these are external ingestion points, the risk is mitigated by the skill's strictly observational nature and the lack of execution tools provided to the agent. No specific boundary markers or sanitization steps are defined for these inputs, but their usage is limited to identifying the target environment for browser-based inspection.

dogfood