---
title: "skill by agenticluke · skilld"
canonical_url: "https://skilld.dev/gh/agenticluke/continuous-agent-loop-plus"
meta:
  description: "Run a steady agent loop with clear quality checks, tests, stop rules, and safe recovery steps. From agenticluke/continuous-agent-loop-plus."
  "og:description": "Run a steady agent loop with clear quality checks, tests, stop rules, and safe recovery steps. From agenticluke/continuous-agent-loop-plus."
  "og:title": "skill by agenticluke"
  "twitter:description": "Run a steady agent loop with clear quality checks, tests, stop rules, and safe recovery steps. From agenticluke/continuous-agent-loop-plus."
  "twitter:title": "skill by agenticluke"
---

`

[All skills](https://skilld.dev/skills)

[![agenticluke avatar](https://skilld.dev/_img/avatar?url=https%3A%2F%2Fgithub.com%2Fagenticluke.png%3Fsize%3D96)](https://skilld.dev/gh/agenticluke)

# **/skill**

[@6670990](https://github.com/agenticluke/continuous-agent-loop-plus/commit/6670990aab468ce236497822fc227efcd26e6176 "Your agent reads SKILL.md at commit 6670990")

by [agenticluke](https://skilld.dev/gh/agenticluke)· [agenticluke](https://skilld.dev/gh/agenticluke)/ [continuous-agent-loop-plus](https://skilld.dev/gh/agenticluke/continuous-agent-loop-plus)

Run a steady agent loop with clear quality checks, tests, stop rules, and safe recovery steps.

- 1 file
- 3.6 KB
- Updated 4 weeks ago
- [GitHub](https://github.com/agenticluke/continuous-agent-loop-plus/blob/main/skill/SKILL.md "View SKILL.md on GitHub")

## SKILL.md

3.6 KB

**≈25** tokens always: the name and description. **≈876** when used: this file.

## Continuous Agent Loop

### Credit

Original pattern by **ECC**. This version keeps the same goal and gives clear steps.

### Purpose

Use this skill for work that needs more than one agent pass.

Each pass must:

1. Do one small task.
2. Check the result.
3. Record clear progress.
4. Choose the next task or stop.

This is the standard loop name for v1.8 and later. Keep `autonomous-loops` as an old alias for one release.

### Pick a Loop

```
Start
  |
  +-- Need strict CI or PR checks? -- yes --> continuous-pr
  |
  +-- Need to split an RFC into tasks? -- yes --> rfc-dag
  |
  +-- Need many test ideas at once? -- yes --> infinite
  |
  +-- Otherwise --> sequential
```

Use `sequential` when you are not sure. It is the safest choice.

### Before You Start

Write down:

- The goal
- What is in scope
- What is out of scope
- How to test the work
- What counts as done
- The retry limit
- The time or cost limit
- When a person must step in

Do not start if the goal or done rules are not clear.

### Loop Steps

Repeat these steps:

1. Pick the smallest useful task.
2. Make the change.
3. Run the needed checks.
4. Compare the result with the done rules.
5. Save what changed, what passed, and what failed.
6. Continue only if there is clear progress.

Stop when:

- All done rules pass.
- The retry limit is hit.
- The time or cost limit is hit.
- The same root cause fails twice.
- A needed tool, file, or right is missing.
- The next step may harm data or users.
- A person must make a choice.

Never claim success when a check was skipped.

### Suggested Stack

For a full work flow, use:

1. RFC task split with `ralphinho-rfc-pipeline`
2. Code checks with `plankton-code-quality` and `/quality-gate`
3. Result checks with `eval-harness`
4. Saved session state with `nanoclaw-repl`

If one tool is not present, note it. Use a local check that tests the same rule. Do not add new web calls.

### Common Failures

- The loop runs but makes no clear progress.
- The same root cause is tried again and again.
- A check gives a false pass due to old files or cached data.
- Tests pass, but the main done rule still fails.
- Two tasks edit the same file and lose work.
- The merge queue stops.
- A tool or service is down.
- A task waits for a person but keeps running.
- Time or cost grows with no limit.
- The loop changes work outside its scope.
- The loop ends without saving its state.

### Recovery

When a failure repeats:

1. Stop the loop.
2. Save the last good state.
3. Run `/harness-audit`.
4. Find the first failed check.
5. Cut the scope to that one failed unit.
6. Clear stale output or cached data if it is safe.
7. Set clear pass rules.
8. Replay once.
9. If it fails for the same root cause, stop and ask for help.

Do not keep retrying with small wording changes. Change the plan, fix the cause, or stop.

### Usage Example

Goal: Add a CSV export button.

Done rules:

- The button is shown on the report page.
- It downloads a valid CSV file.
- Empty reports still download a file with headers.
- Commas, quotes, and new lines are escaped.
- Current tests pass.
- One new export test passes.
- Retry limit is two.

Loop:

```
Pass 1:
  Add the export code.
  Run the export test.
  Result: Failed. Quotes are not escaped.

Pass 2:
  Fix CSV escaping.
  Run the export test and all current tests.
  Result: All checks pass.

Stop:
  Save the result.
  Mark the goal done.
```

If Pass 2 fails for the same quote bug, stop the loop. Save the error and use the recovery steps.

Source: [SKILL.md on GitHub](https://github.com/agenticluke/continuous-agent-loop-plus/blob/main/skill/SKILL.md)

## Third-party checks

No third-party reports yet.

## Provenance

[Signed by skilld at 6670990.](https://github.com/agenticluke/continuous-agent-loop-plus/commit/6670990aab468ce236497822fc227efcd26e6176 "6670990aab468ce236497822fc227efcd26e6176") This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 4 weeks ago.

Activeupdated 4 weeks ago

## Capability

<dl>

<dt>origin</dt>
<dd>ECC</dd>

</dl>

## README badge

![README badge for agenticluke/continuous-agent-loop-plus](https://skilld.dev/b/agenticluke/continuous-agent-loop-plus?theme=light&label=0)

## Related skills

-
-
-
-
-
-