verify-task-result

name	verify-task-result
description	Create a verifying agent to check whether a working agent's task was completed successfully. Use after a working agent finishes (transitions to "done", "stopped", "waiting", or "paused") and appears to believe its task is complete.

Verifying a task result

Every completed task must be verified. The verifying agent independently reviews what was done, judges whether it satisfies the original intent, and produces a concrete set of recommended next actions for you to execute.

You should not try to gather or summarize the working agent's output yourself -- that would pollute your context. Instead, give the verifying agent all the pointers it needs to investigate the work on its own.

What to include in the verification task message

Write a markdown file for the verifying agent that includes:

1. The working agent's ID so the verifier can inspect its transcript, artifacts, and state directly (e.g., via mngr transcript <working-agent-id>)
2. The original task description -- what you asked the working agent to do, and why. Copy or closely paraphrase the original --message you sent when creating the worker. Include the success criteria.
3. Any follow-up context -- if you sent additional messages to the working agent during its run (e.g., answering its questions, providing clarifications, updating requirements), mention that you did so and what the key points were. The verifier can read the full transcript, but knowing what changed helps it focus.
4. The user's original request (if applicable) -- if this task was triggered by a user message, include any relevant conversation IDs and time-spans for each so the verifier can judge whether the result actually satisfies the user's intent (not just the task instructions as you interpreted them).
5. High level purpose -- if the task is part of a larger project (and especially if this was not a task that the user asked for directly), include a brief description of the overall purpose or goal of the project and how the task fits in.