| name | mission-review-loop |
| description | Use when a completed or terminated mission enters human review and Squad must verify mission and task outcomes, surface evidence, collect feedback, and preserve learning without treating human attention as missing machine proof. |
Mission Review Loop
Goal
Help the human decide whether the mission delivered the required outcome while
Squad handles all machine-verifiable review work first.
The user-facing relief is:
Squad shows me the outcome, the proof, and the few decisions that actually need
my judgment.
Mission Review is a chat collaboration, not a Mission Overview mode. Mission
Overview can launch review and display mission readback, but the back-and-forth
review process happens in a new mission-scoped Chat with this skill loaded.
Use This Skill When
- A completed, blocked, or terminated mission exposes
Review mission.
- Chat checks mission status and finds a completed or review-ready mission.
- The user opens Mission Review from Mission Control, from the Chat side rail
above conversations, or from Chat after the agent checks mission status.
- Feedback from review should update future mission memory or trigger follow-up
work outside mission runtime.
Do Not Use This Skill When
- The mission is still defining; use
mission-definition-conveyor.
- A task is actively running; use
mission-task-operation.
- The user only wants status without a review decision; use
mission-overview-readmodel.
Required Inputs
- Mission operating context and final mission readback.
- Mission requirements, acceptance criteria, task DAG, authority, proof package,
receipts, artifacts, and review surface.
- Task final artifacts.
- Frontend media, backend endpoint proofs, devops deployment proofs, and
telemetry/runtime-proof receipts as required by task type.
- Human decision question and feedback capture target.
Review Procedure
- From Mission Control, Chat side rail, or Chat status output, route the user
to the Mission Overview route for the selected mission.
- From Mission Overview, the
Review button opens a new chat conversation
scoped to mission review and loads this skill.
- Load mission overview and proof/readiness readback into Chat.
- Review the mission layer: outcome, requirements, acceptance, authority,
task DAG, proof package, and open
HOLDs.
- Review every task layer:
- backend: hit endpoints or replay tests where safe;
- frontend: display screenshots/videos and verify user flows;
- devops: verify health, hashes, release/feed posture, and frontend impact.
- Use generated UI for mission cards and task cards when it helps the human
inspect outcomes, proof media, and task artifacts in the conversation.
- Run autonomous checks before asking the human to judge.
- Surface only irreducible human attention:
- whether the delivered outcome is acceptable;
- whether UX/copy/behavior feels right;
- whether a risk or deferral is acceptable.
- Capture feedback and memory updates for future missions.
- If changes are needed, collaborate in chat directly. Do not reopen the
terminated mission runtime unless a new mission is explicitly created.
Proof Rules
- Backend endpoint proof does not prove frontend outcome.
- Frontend media does not prove durable state.
- Deployment health does not prove product behavior.
- A proof package summary does not prove each task receipt.
- Human approval does not replace missing machine proof.
Output
Return a mission review packet with:
- mission outcome verdict;
- requirement-by-requirement status;
- task verdicts and receipt refs;
- mission card/task card generated UI refs when used;
- proof surfaces satisfied;
- machine-verifiable gaps;
- human judgment decisions;
- feedback captured for memory;
- recommended follow-up path: accept, changes in chat, new mission, or blocker.
Stop Condition
Hold review when:
- proof package is missing for a claim that requires it;
- task final artifacts are absent;
- required frontend media is missing;
- runtime-proof telemetry cannot distinguish happy and negative paths;
- the human would need to reconstruct context the agent should have gathered.
- review is being kept inside Mission Overview instead of being launched into
mission-scoped Chat.
Receipt
Receipt must include:
- mission refs and review surface refs;
- task final artifact refs;
- proof queries and media paths;
- human decision packet;
- memory update or explicit non-update;
- remaining
HOLDs and non-claims.