> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cortex-labs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Review and decide an Explorer plan

> Inspect a read-only plan, request changes or accept implementation, and check the recorded decision.

Use **Plan** mode when you want an Explorer agent to investigate the task and propose its approach before implementation. You need access to the workspace agent chat and permission to update it to decide a plan.

Plan mode lets the agent read, search and ask questions. It authors its plan without changing workspace content. **Build** mode enables the selected agent's full permitted toolset. Accepting a plan switches to Build and starts implementation under your current authority and execution limits.

<Steps>
  <Step title="Enter Plan mode and state the scope">
    Open the Explorer chat. Use the mode toggle in the composer or press **Shift+Tab** to select **Plan**. The agent can also enter Plan mode during a conversation.

    Describe the result, permitted inputs, limits and any questions to resolve. Replace the bracketed fields with your own task details and ask:

    > Plan a short investigation note for \[workspace name] using \[my permitted collection or dataset]. Address \[question] within \[scope and optional exact period with time zone]. Propose how to inspect source versions, compare conflicting claims and record coverage gaps. Save the resulting note as a workspace draft when implemented; leave collection, processing and publication unchanged.

    Answer clarification questions. When the plan is ready, the agent hands it over and waits for a decision.
  </Step>

  <Step title="Inspect the newest handoff">
    Open the **Plan** tab and read the newest plan. Check the intended inputs, evidence checks, proposed changes, output and stopping point. Resolve uncertain access, data readiness or tool effects before accepting.

    If the chat has earlier plans, use the picker above the plan to inspect them. Only the newest handoff can be decided or annotated. Return to it before taking action.
  </Step>

  <Step title="Make the decision in the plan controls">
    Choose the outcome that matches your intent:

    | Control | Result |
    | - | - |
    | **Accept** | Records acceptance, switches the chat to Build and starts work on the plan. |
    | **Reject** | Records that the plan was not approved, keeps Plan mode and starts no new turn. |
    | **Review** | Sends your comments for revision, keeps Plan mode and starts a new planning turn. |

    To request changes, add comments against the relevant plan lines, then send them with **Review**. Make the required adjustment explicit, such as excluding a dataset or narrowing the output. Inspect the new handoff before deciding it.

    Typing a message such as “looks good” in the chat does not settle a pending plan. Use the plan decision controls; an ordinary chat message can start another turn while the handoff remains pending.
  </Step>

  <Step title="Verify the decision and resulting work">
    Check the transcript for **Plan accepted**, **Plan not approved** or **Plan sent back for revision**. An accepted newest plan also shows **Plan accepted** above it in the Plan tab.

    After acceptance, inspect the actual work status and resulting saved output. Acceptance records permission to implement; it does not prove that execution completed or that every requested operation succeeded.

    After Review, check that a revised plan addresses your comments. After Reject, confirm that implementation did not start from that decision. Earlier plans remain available for comparison; deciding the newest plan does not rewrite them.
  </Step>
</Steps>

## Check the result

The newest handoff should have a recorded decision that matches your intent. If accepted, the chat should be in Build and you should know what has completed and what remains unfinished.

Plan acceptance is separate from review and publication of an artefact. A resulting output still needs approval of its selected version before publication. Plain Arc chats and workflow Agent steps do not use this Explorer plan-decision path; a workflow's human decision belongs to its configured review step.

Continue with [Investigate in Explorer](/vantage/explore-evidence/investigate) and [Follow a finding to its evidence](/vantage/explore-evidence/follow-evidence). For a repeatable process with a human reviewer, see [Build and test a workflow with Kato](/kato/build-workflow).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.