Rinari docsWorking with Rinari
PLAN, BUILD and REVIEW
Choose how Rinari approaches a task. Each mode runs under its own engine-enforced permissions, not a different prompt.
You are here
The mode tells Rinari how to approach the task, and the engine applies the permissions that go with it. A mode is an execution rule the engine enforces, not a tone of voice. Switching modes keeps the session, its tasks and its context.
Choose the mode in the composer, next to the model.
PLAN
PLAN investigates and proposes; it doesn’t touch the project.
- Reads files and the internet freely.
- Never modifies files and doesn’t run commands.
- Ends with a plan: relevant code, risks, ordered steps and how to validate them.
When a plan is ready, Rinari asks Implement this plan?:
- Implement plan continues in BUILD in the same conversation, implements the plan and verifies the changes.
- Not now leaves it as a plan.
BUILD
BUILD is the working mode: Rinari edits files, runs commands and tests, uses Git and delegates to subagents — within the permissions you chose for the chat:
| Permissions | Behaviour |
|---|---|
| Read-only | Reads files and the internet. Never writes, runs or sends anything. |
| Workspace | Free within the project, localhost and the internet for reading. Asks once to act outside: writing, sending data, git push. |
| Full access | Doesn’t ask. Only force push, deleting outside the project and system keys ask for confirmation. |
See Approvals and permissions for exactly what asks and how to answer, and Full access for the confirmation it asks before turning on.
REVIEW
REVIEW reads and reports; it doesn’t implement.
- Reads files and the internet freely.
- Never modifies files.
- Asks for your approval before each command it wants to run — for example, to run the test suite.
- Reports findings about the current changes, architecture, tests, security or regressions.
Reading scope in PLAN and REVIEW
PLAN and REVIEW always run with an execution profile that denies writes and shell execution. The permission selector still matters: it decides how far they can read.
| Selector shows | Reads |
|---|---|
| Read · Full access | Any file, without asking. |
| Read · Workspace | Inside the project freely; asks before reading outside it. |
| Read-only | Only inside the project. |
Sensitive credentials — system keys and credential stores — keep their own approval in every mode.
In the terminal
The CLI has a command for each approach:
rinari plan "Add JSON export to the report command"
rinari agent "Fix the failing auth tests"
rinari review "Check the current diff for regressions"ask, plan and review use a read-only profile; implementation commands use a workspace profile. In the terminal these are separate task invocations, not an automatic plan-to-build pipeline: review the result and decide the next step.
See how a conversation moved between modes
The Flows view draws a conversation as stages — Planning, Implementation, Review — so you can see when it planned, built and reviewed. See Boards and Flows.