plan-mode-permission-gating

SKILL: Plan Mode Permission Gating

Domain: permission-gating Trigger: Invoke when the user wants to enter plan mode or a complex implementation requires additional permission scrutiny before introducing file edits. Source Pattern: Distilled from reviewed plan-mode entry and permission-transition implementations.

Core Method

Wrap plan-mode entry in an explicit permission transition. Validate that the request comes from the root session, switch the shared permission context into a planning state, and emit a standardized banner that reminds agents to explore before editing. Centralizing that transition prevents individual call sites from bypassing the plan-mode contract or drifting into edits before planning is complete.

Key Rules

• Treat plan-mode entry as a read-only permission change that cannot be invoked inside a child-agent context.
• Reconfigure the shared permission context through one centralized transition path so state sync and side effects stay consistent.
• Return a human-facing banner describing the plan-mode obligations, and let follow-up instructions adapt to the active plan-mode variant.
• Keep the tool’s UI payload minimal but explicit about the no-edit rule; reuse the same instructions for both regular and interview-enhanced plan mode.

Example Application

Use this skill when a CLI state machine detects a non-trivial refactor request: enter plan mode first, flip the session into a no-edit planning state, and require a reviewed plan before touching files.

Anti-Patterns (What NOT to do)

• Don’t let a code path change permissions directly without the standardized transition, or some agents will edge into edits without the plan-mode reminder.
• Don’t ignore agent forks — the gate belongs only to the root session, so refuse plan-mode tool calls when context.agentId is set.