Writ by Infinri
Writ and native Claude Code
Claude Code already provides permissions, plan mode, hooks and memory. Writ packages a particular approval workflow and decision-recording system on those extension points.
What each provides
This comparison describes documented capabilities. It is not a controlled performance test. Writ's side is pinned to commit 8cd4f81c7a22; native documentation was checked October 25, 2026.
| Need | Native Claude Code | Writ |
|---|---|---|
| Control actions | Permission rules, modes and hooks can allow, deny or request approval. | Uses native hooks to apply its defined checks to supported actions. |
| Review before editing | Plan mode supports exploration and plan review before implementation. | Work mode adds a staged plan checkpoint and a separate test-skeleton checkpoint, with documented path exemptions. |
| Bind approval to the plan | Hooks let you implement custom workflow checks. | Maintains approval state tied to the reviewed plan's fingerprint and checks for drift. |
| Supply project context | Instruction files, scoped rules and auto memory provide context. | Retrieves relevant rules and documents and adds file-specific context at supported write points. |
| Remember why files changed | Auto memory preserves selected project knowledge between sessions. | Records selected approved decisions and commit/file reasons for project-scoped Recall. |
| Install and maintain | Built-in features and any hooks you choose to configure. | Adds a Python environment, a local daemon, Docker and Neo4j, plus Writ's configuration and updates. |
When native features may be enough
If you need a few permission rules, concise instructions or one blocking hook, start with the native tools. They may cover the task with less setup. Writ does not need to be installed to create a Claude Code hook.
When to evaluate Writ
Evaluate Writ if you want its defined plan and test acceptance sequence, approval binding, relevant rule retrieval and decision provenance together. Try the approval-gate walkthrough and assess whether the workflow is worth the additional services for your project.
Keep the security boundaries clear
Writ is not a security sandbox. Its gates constrain supported actions by a cooperative agent, and its inspection has documented gaps. Native permission controls and sandboxing have their own scope. Do not treat this comparison as proof that either setup provides complete containment.