The Agent Skills Directory

[SAFE]: The skill implements an autonomous validation workflow that relies on standard git commands and internal sub-agent orchestration. The stated purpose of multi-agent verification matches the implementation logic.
[COMMAND_EXECUTION]: The skill uses git diff to identify modified files and summarized change statistics. This is restricted to metadata gathering and does not involve executing arbitrary shell commands or untrusted external scripts.
[DATA_EXPOSURE]: The workflow explicitly instructs the agent to avoid reading full file contents into the orchestrating context window, which minimizes the risk of sensitive data exposure during the orchestration phase. It uses path-based references for specs and architecture documents.
[INDIRECT_PROMPT_INJECTION]: As a multi-agent orchestrator, the skill has an inherent attack surface for indirect prompt injection because it processes feedback (JSON findings) from other agents. However, the risk is mitigated by the modular design where specialized agents perform the actual file analysis in separate contexts, and the orchestrator acts on structured recommendations rather than executing raw text from external sources.

validation-loop