Run agents, automations and loops

Answer a loop exception and correct repeated mistakes

A loop exception is a case that needs information, a person's decision or attention after a failed check. A correction records how an output should change. Use those specific cases to fix repeated mistakes in a recurring workflow, while keeping the evidence behind the change.

Contents

A loop is the workflow definition; a case is one occurrence of it. Answering one case and revising the definition are separate jobs. You may need to do both when the same problem returns.

Deal with the current request first

Open the item in your Portal's Inbox and inspect the linked case timeline. Read the question or failed check, the proposed result and the source evidence. Use the request's offered answer or review controls within your authority.

Keep the original reference if the response is interrupted. Inspect that request and case before answering again. A successful answer does not prove that the next step ran; return to the case and check its new state.

Keep the useful correction

An edited proposal says what the output should have been. A rejected proposal says that the original output was wrong, but may not establish a single correct replacement. Keep that distinction when describing the error to your agent.

If repeated errors matter, pause or reduce the workflow's autonomy through the authorised controls while you investigate. Autonomy, budgets and access explains why a demotion applies to the relevant action rather than granting or removing every permission at once.

Draft a replacement from the recorded examples

Ask your connected agent to use the loop-reviser skill. It reads the current definition and the saved corrections, names the recurring pattern and proposes the smallest useful replacement. The new version names the current loop as its predecessor and starts in shadow mode.

The correction examples are also regressions: cases the replacement must handle without reproducing the problem. The author needs a supported operator route to inspect that evidence and run replay. If the connected client cannot do that, hand over the loop reference, case references and intended correction rather than claiming a replay passed.

Read the replay result before approval

Replay substitutes recorded action outputs instead of invoking those actions. Its quality checks still run and can incur model cost. Inspect the result for every saved case; an unreadable or ungradable example is not a passing result.

An authorised person reviews the exact replacement and its replay result before approval. Activation is a further decision. Approval can be refused when the required replay has not passed. Do not use a waiver simply to get past a failing correction; resolve the cause or leave the revision waiting with a clear explanation.

Use corrections to check a replacement

A replay result precedes the owner’s decision.

Read the existing case corrections and identify the recurring problem. Compose the smallest successor in shadow mode. Replay the saved regressions: actions use recorded outputs while checks run for real and can incur cost. An authorised person reviews the exact candidate and replay result. Approval and later activation remain separate; an ungradable regression is not a pass.

Related

Understand a loop case and follow its result · Build your first recurring workflow with a loop · Control agent autonomy, budgets and access · Review what needs your attention

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.

---

# Answer a loop exception and correct repeated mistakes

Guides · Run agents, automations and loops · [How it works](https://app.mensara.ai/how-it-works)

A loop exception is a case that needs information, a person's decision or attention after a failed check. A correction records how an output should change. Use those specific cases to fix repeated mistakes in a recurring workflow, while keeping the evidence behind the change.

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

[Operating procedure](https://app.mensara.ai/how-it-works/loop-correction-procedure)

A loop is the workflow definition; a case is one occurrence of it. Answering one case and revising the definition are separate jobs. You may need to do both when the same problem returns.

## Deal with the current request first

Open the item in your Portal's **Inbox** and inspect the linked [case timeline](https://app.mensara.ai/how-it-works/loop-cases). Read the question or failed check, the proposed result and the source evidence. Use the request's offered answer or review controls within your authority.

Keep the original reference if the response is interrupted. Inspect that request and case before answering again. A successful answer does not prove that the next step ran; return to the case and check its new state.

## Keep the useful correction

An edited proposal says what the output should have been. A rejected proposal says that the original output was wrong, but may not establish a single correct replacement. Keep that distinction when describing the error to your agent.

If repeated errors matter, pause or reduce the workflow's autonomy through the authorised controls while you investigate. [Autonomy, budgets and access](https://app.mensara.ai/how-it-works/control-agent-autonomy) explains why a demotion applies to the relevant action rather than granting or removing every permission at once.

## Draft a replacement from the recorded examples

Ask your connected agent to use the **loop-reviser** skill. It reads the current definition and the saved corrections, names the recurring pattern and proposes the smallest useful replacement. The new version names the current loop as its predecessor and starts in shadow mode.

The correction examples are also **regressions**: cases the replacement must handle without reproducing the problem. The author needs a supported operator route to inspect that evidence and run replay. If the connected client cannot do that, hand over the loop reference, case references and intended correction rather than claiming a replay passed.

## Read the replay result before approval

Replay substitutes recorded action outputs instead of invoking those actions. Its quality checks still run and can incur model cost. Inspect the result for every saved case; an unreadable or ungradable example is not a passing result.

An authorised person reviews the exact replacement and its replay result before approval. Activation is a further decision. Approval can be refused when the required replay has not passed. Do not use a waiver simply to get past a failing correction; resolve the cause or leave the revision waiting with a clear explanation.

### Use corrections to check a replacement

A replay result precedes the owner’s decision.

Read the existing case corrections and identify the recurring problem. Compose the smallest successor in shadow mode. Replay the saved regressions: actions use recorded outputs while checks run for real and can incur cost. An authorised person reviews the exact candidate and replay result. Approval and later activation remain separate; an ungradable regression is not a pass.

## Related

- [Understand a loop case and follow its result](https://app.mensara.ai/how-it-works/loop-cases)
- [Build your first recurring workflow with a loop](https://app.mensara.ai/how-it-works/first-loop)
- [Control agent autonomy, budgets and access](https://app.mensara.ai/how-it-works/control-agent-autonomy)
- [Review what needs your attention](https://app.mensara.ai/how-it-works/review-attention)

Previous page: [Understand a loop case and follow its result](https://app.mensara.ai/how-it-works/loop-cases)

Next page: [Control agent autonomy, budgets and access](https://app.mensara.ai/how-it-works/control-agent-autonomy)

Autonomy is how much a workflow may do without stopping for a person.

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