Plan Mode

Plan mode is a read-only execution mode in Claude Code that separates the exploratory, analytical phase of a task from the implementation phase. In plan mode, the model can read files, search the codebase, run read-only commands, and reason about what changes would be needed — but it cannot write files, edit code, or execute commands with side effects. The session remains in this constrained state until you explicitly exit plan mode and approve the proposed approach.

The value of this separation is architectural clarity under uncertainty. Complex refactoring tasks, system migrations, and multi-file changes involve a set of interdependent decisions: which files need to change, in what order, what the ripple effects are, and where the risks lie. Attempting all of this in a single pass — reading, deciding, and writing simultaneously — means committing to implementation details before the full picture is understood. Plan mode forces a deliberate pause: explore first, decide second, implement third.

From a team workflow perspective, plan mode creates a natural review checkpoint. The model produces a written plan — a list of proposed changes, the reasoning behind each, and the risks it identified — which a human can review and refine before any files are touched. This is particularly valuable in codebases where a wrong assumption early in a refactor can cascade into hours of untangling.

Plan mode also interacts with skills and subagents. A skill can be authored to enter plan mode automatically for tasks that match certain patterns (e.g., any task that involves schema changes or public API modifications), giving teams a repeatable workflow that enforces review for high-stakes operations. Pairing plan mode with iterative-refinement — where each planning cycle tightens the proposal before committing — is a common advanced pattern for large-scale changes.

The context: fork option and plan mode address different problems and are often combined. context: fork handles context pollution; plan mode handles premature commitment to implementation. A skill that uses both runs its entire analysis in an isolated context and returns a clean plan without modifying anything — the ideal starting point for a careful change.

For the CCA-F exam, the key distinction is between plan mode (enforces read-only behavior at the session level) and context: fork (isolates context at the skill level). They can coexist but serve different purposes. The Plan Mode vs Direct Execution concept page covers the tradeoffs that exam questions draw on when presenting scenarios where the wrong mode choice leads to avoidable rollbacks or missed edge cases.

Exam questions typically present a scenario — a large refactor, an ambiguous requirement, a high-stakes migration — and ask which Claude Code feature should be used before implementation begins. Plan mode is the answer when the concern is "making changes before the full picture is understood." Distractor answers often include context: fork (wrong layer) or structured output flags (unrelated to execution mode).