Prinevo.aiDocs
Docs/Run work

Run work

Run one specialist in single-node mode

Single-node mode gives one specialist a focused task while keeping the run, context, artifacts, and result auditable.

Product label Start single-node mode by choosing Direct agent or mentioning a specialist in the composer.
Current product guideUpdated October 28, 2026

When to use it

Use a single-node run when one agent owns the outcome and you do not need the complete Product-to-Learning sequence. Common examples include architecture review, pull-request review, focused verification, release preflight, or a bounded analysis.

Start a run

Type @ in the composer to open the specialist picker shown below.

  1. 1
    Start a new session

    Open Sessions and select New.

  2. 2
    Select Build

    Direct agent is a build path even when the result is an analysis or review artifact.

  3. 3
    Mention the specialist

    Type @ and select an agent such as Architect, Code Review, or Verify.

  4. 4
    Add the target and outcome

    Select repository hints and include a pull-request URL or other bounded target when relevant.

  5. 5
    Submit

    Prinevo creates one durable single-node run owned by the selected specialist.

Type @ in Build and choose Product, Architect, Lead, Implement, Code Review, or Verify.
Type @ in Build and choose Product, Architect, Lead, Implement, Code Review, or Verify. Captured in the product; states reflect the capture, not your workspace.Open full size

Agent selection and lane selection are different. Typing @ selects one specialist and starts Direct agent work in single-node mode. To run Inline quick, Quick Agent, or Full lane, use the Build lane controls or the corresponding lane command.

QA example: acceptance criteria, scenarios, screenshots, and video · Code Review example and dimensions · Read the resulting artifacts

Examples

@Verify validate pull request 786. Run the relevant tests, inspect the changed contracts, and return the evidence needed to approve or block the merge.
@Architect review the proposed billing boundary. Identify cross-service risks and produce a publishable architecture recommendation.

Review the result

  1. Open the specialist’s latest artifact and confirm it answers the requested question.
  2. For PR work, confirm the repository, PR URL, branch, and revision match your target.
  3. Inspect findings and check results. A review artifact alone does not prove tests passed.
  4. Resolve blocking findings or missing access before treating the task as complete. Give feedback in the same session; use the available retry action for a failed attempt.
  5. Choose the next action explicitly: revise the change, run another specialist, or start the appropriate delivery lane.
QA report links and the attached screenshot, video, and supporting-evidence groups.
QA report links and the attached screenshot, video, and supporting-evidence groups. Captured in the product; states reflect the capture, not your workspace.Open full size

Single-node work completes the selected specialist’s task. It does not automatically run the remaining SDLC stages or authorize merge and deployment.

Use a specialist to review or verify a pull request · Recover a failed run

What stays controlled

The selected agent still uses its saved model, connection, reasoning level, compute profile, skills, and tool policy. The run keeps its own status, artifacts, decisions, and completion contract. If the work grows into a multi-stage feature, start or switch to Full lane instead of treating a single specialist as an entire delivery team.