safety-modes

What problem does it solve?

It reduces the chance of unintended damage by forcing elevated-risk work into an explicit, operator-visible safety workflow before anything risky happens.

Core Features & Use Cases

• Mode-based safety gating: Supports careful, freeze-scope, and investigate-first to match different risk profiles and uncertainty levels.
• ProposedAction risk/result tracking: Requires each risky or boundary-sensitive step to be recorded with ActionRisk and an ActionResult lifecycle (planned, blocked, approved, applied, rolled-back, abandoned).
• Freeze boundary enforcement: In freeze-scope, constrains edits to an allowed boundary and includes lock-conflict verification for boundary files to prevent contention-related failures.
• Evidence-first investigation: In investigate-first, separates evidence from assumptions and blocks fixes until hypotheses are explicit and testable.
• Strict-safety compatibility: When enabled, treats the checklist as a decision gate and prevents destructive actions without recorded explicit approval, halting on policy violations.

Quick Start

Enter the requested mode for your current work session by asking the agent to produce a safety checklist for the risky task you intend to perform, including proposed actions with ActionRisk and ActionResult.