For agents and technical readers

Prepare a loop correction and replay its replacement

Turn recorded corrections into a proposed replacement. Keep composition, replay, approval and activation as separate observed stages.

Contents

Read the actual failure

Use get_loop and get_case to inspect current definition, case timeline, regression count and replay verdict. Load get_skill with {"name":"loop-reviser"}. Its author needs a supported route to the saved regression evidence; a count alone is not the content of that evidence.

If the client cannot inspect regressions or execute replay, hand the loop and case references to an authorised operator with that route. Do not invent a public get_regressions tool or claim replay from schema inspection.

Compose the minimum replacement

Name the repeated pattern from the evidence. An edited output supplies a positive expected correction; rejection only establishes that the earlier output was wrong. Propose the minimum change using loop_compose with the full current supported LoopSpec, a higher version, supersedes naming the current loop id and shadow rollout. Use the live action palette and existing loop composition procedure.

Do not widen triggers, effects or record access without the user's authority. Read and retain the newly returned loop id.

Replay and inspect the exact candidate

Where current act exposes the authorised replay action:

{"action":"loop_replay","item_id":"<new loop id>"}

Inspect per-case results, not only an overall label. Replay substitutes recorded action outputs, while checks run for real and can incur model cost. corrected-output-not-produced and drafted-output-reproduced mean the candidate still fails the correction. no-recording, unreplayable, unreadable or chain-truncated mean that case was not graded; none is a pass.

Hand over for a decision

Report new loop id, predecessor, version, changed behaviour, regressions replayed and failures. An authorised person approves through the supported loop_approve action only after inspecting the exact candidate and result. Approval may require the passing replay; activation is separate. Do not suggest an owner waiver as a way around an unresolved failing replay.

After approval/activation, inspect a normal new case. Passing a replay does not prove that a scheduled case ran with live inputs. Preserve old case history and reconcile any already-open case before assuming it moved to the new version.

Practical guides for people and agents

Start with one useful job

Join the waitlist

Plain text for AI agents. Same content as the human page.

## For agents

- What Mensara is: it gives a team its own Mind, one shared memory for the team and their AI agents, built by Xanadu Labs LLC.
- Status: signup is not open. A person joins the waitlist at https://app.mensara.ai/#waitlist. Agents cannot create an account or sign up.
- How you can help your person: explain what Mensara does, check whether it fits a job their team repeats, and share the waitlist link.
- Agent references: https://app.mensara.ai/llms.txt lists every page as Markdown and the tool input references for connected agents.

---

# Prepare a loop correction and replay its replacement

Operating procedures · For agents and technical readers · [How it works](https://app.mensara.ai/how-it-works)

Turn recorded corrections into a proposed replacement. Keep composition, replay, approval and activation as separate observed stages.

[Page as Markdown](https://app.mensara.ai/how-it-works/loop-correction-procedure.md)

[Human guide](https://app.mensara.ai/how-it-works/correct-a-loop)

## Use the released interface with current authority

Availability: Core0.20.11 / Portal0.3.47 source checked; discover the actual installed client and enabled features.

Who can do this: The connected principal with the permissions named in this procedure

What changes: The procedure explains its actual reads, writes or external effects before action.

Limits: An input accepted by a source schema is not a successful installed-runtime task.; Keep human decisions, client exclusions and pending results explicit.

Verification: Source checked and independently reviewed; scoped runtime receipts are recorded separately.

## Read the actual failure

Use `get_loop` and `get_case` to inspect current definition, case timeline, regression count and replay verdict. Load `get_skill` with `{"name":"loop-reviser"}`. Its author needs a supported route to the saved regression evidence; a count alone is not the content of that evidence.

If the client cannot inspect regressions or execute replay, hand the loop and case references to an authorised operator with that route. Do not invent a public `get_regressions` tool or claim replay from schema inspection.

## Compose the minimum replacement

Name the repeated pattern from the evidence. An edited output supplies a positive expected correction; rejection only establishes that the earlier output was wrong. Propose the minimum change using `loop_compose` with the full current supported LoopSpec, a higher version, `supersedes` naming the current loop id and shadow rollout. Use the live action palette and existing [loop composition procedure](https://app.mensara.ai/how-it-works/build-loop-procedure).

Do not widen triggers, effects or record access without the user's authority. Read and retain the newly returned loop id.

## Replay and inspect the exact candidate

Where current `act` exposes the authorised replay action:

```
{"action":"loop_replay","item_id":"<new loop id>"}
```

Inspect per-case results, not only an overall label. Replay substitutes recorded action outputs, while checks run for real and can incur model cost. `corrected-output-not-produced` and `drafted-output-reproduced` mean the candidate still fails the correction. `no-recording`, `unreplayable`, `unreadable` or `chain-truncated` mean that case was not graded; none is a pass.

## Hand over for a decision

Report new loop id, predecessor, version, changed behaviour, regressions replayed and failures. An authorised person approves through the supported `loop_approve` action only after inspecting the exact candidate and result. Approval may require the passing replay; activation is separate. Do not suggest an owner waiver as a way around an unresolved failing replay.

## Related

- [Answer a loop exception and correct repeated mistakes](https://app.mensara.ai/how-it-works/correct-a-loop)
- [Read a loop case and reconcile a wait](https://app.mensara.ai/how-it-works/inspect-loop-case-procedure)
- [Control agent autonomy, budgets and access](https://app.mensara.ai/how-it-works/control-agent-autonomy)

[Join the waitlist](https://app.mensara.ai/#waitlist)