implementation-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides a set of structured guidelines for an AI agent to manage repository edits safely. Key security features include:
- Least Privilege and Evidence-Based Action: Instructions mandate reading local project conventions and inspecting the repository before taking action or asking questions.
- Separation of Planning and Execution: The skill enforces 'Plan only' modes and requires plan approval before file modification.
- Integrity Protection: Explicit rules prevent overwriting user work, require inspecting current diffs before editing overlapping files, and prohibit marking tasks as complete if changes cause check failures.
- Secure State Management: Temporary task state (plans) is stored in the system temp directory and deleted upon completion, avoiding persistent pollution of the repository.
- Safe Tool Use: Guidance on subagents emphasizes using them for read-only research and risk analysis, with an orchestrator handling integration to avoid overlapping edits.
Audit Metadata