lark-drive Rapport d’audit.

Avertissements

• Dependency or permission surface needs review
• Permission surface may require sandboxing
• The SKILL.md depends on ../lark-shared/SKILL.md for authentication, permissions, and safe execution rules, but that file is outside the submitted skill directory. If the skill bundle does not include lark-shared, the instructions will be incomplete at runtime.
• The provided excerpt is truncated at the end, making it unclear whether the full SKILL.md includes all remaining routing rules and safety boundaries; the visible content is otherwise thorough.
• Several high-risk operations are documented as requiring explicit user confirmation and --yes, but some shortcut descriptions still allow broad natural-language triggers that require the agent to correctly decide when to stop. This is mitigated by the workflow references, but remains an operational risk if those references are not read.
• Quality score needs review
• Permission surface needs review: secrets or environment access, shell or command execution
• Dependency/runtime risk: command execution surface, credential or environment access
• Permission surface: secrets or environment access, shell or command execution

Méthode

This report combines public metadata, AI review output, repository freshness, install readiness, OpenAgentSkill events, quality scoring, trust checks, and the agent safety gate. It is not a full source-code security review.