Initial release of openclaw-auto-continue skill
This commit is contained in:
Alice (OpenClaw)
BIN
openclaw-auto-continue.skill
Normal file
BIN
openclaw-auto-continue.skill
Normal file
Binary file not shown.
142
openclaw-auto-continue/SKILL.md
Normal file
142
openclaw-auto-continue/SKILL.md
Normal file
@@ -0,0 +1,142 @@
|
||||
---
|
||||
name: openclaw-auto-continue
|
||||
description: Use when an OpenClaw user wants auto-next, auto-continue, auto follow-up, or wants the agent to continue multi-step work in the same chat without waiting after every step.
|
||||
---
|
||||
|
||||
# OpenClaw Auto Continue
|
||||
|
||||
## Overview
|
||||
Use this skill when an OpenClaw agent should keep going in the same session instead of stopping after each intermediate step.
|
||||
|
||||
This feature is exposed as the chat command:
|
||||
- `/auto-continue`
|
||||
- alias: `/ac`
|
||||
|
||||
It is a **session-scoped switch**, not a global default.
|
||||
|
||||
## When to Use
|
||||
Use when the user says things like:
|
||||
- "開 auto next"
|
||||
- "自動續跑"
|
||||
- "不要每一步都停下來"
|
||||
- "auto continue"
|
||||
- "auto next"
|
||||
- "checkpoint 後自己繼續"
|
||||
|
||||
Do **not** use it when the user wants explicit approval at each step, or when the work contains risky / destructive decisions that should pause for confirmation.
|
||||
|
||||
## Commands
|
||||
### Enable
|
||||
```text
|
||||
/ac on
|
||||
```
|
||||
Equivalent full form:
|
||||
```text
|
||||
/auto-continue on
|
||||
```
|
||||
|
||||
### Disable
|
||||
```text
|
||||
/ac off
|
||||
```
|
||||
Equivalent full form:
|
||||
```text
|
||||
/auto-continue off
|
||||
```
|
||||
|
||||
### Status
|
||||
```text
|
||||
/ac status
|
||||
```
|
||||
Equivalent full form:
|
||||
```text
|
||||
/auto-continue status
|
||||
```
|
||||
|
||||
## Effective Behavior
|
||||
When enabled, the current session is configured to:
|
||||
- `queueMode = "followup"`
|
||||
- `queueDebounceMs = 0`
|
||||
- `queueCap = 1`
|
||||
- `queueDrop = "summarize"`
|
||||
|
||||
When disabled, the session clears these queue settings and clears queued items.
|
||||
|
||||
## Important Boundaries
|
||||
- This is **per session / per chat**, not global.
|
||||
- It does **not** mean the agent should ignore approvals, safety boundaries, or destructive-action confirmation.
|
||||
- It does **not** guarantee infinite autonomous looping; it mainly makes follow-up work in the same session continue more aggressively.
|
||||
- If the user wants it active in a specific chat, enable it in that chat.
|
||||
|
||||
## How to Explain It to Users
|
||||
Short version:
|
||||
- "開了之後，我會比較傾向在同一串裡自動接下一步，不會每一步都停住等你推。"
|
||||
|
||||
Practical version:
|
||||
- Good for multi-step execution
|
||||
- Good for checkpoint-then-continue behavior
|
||||
- Not ideal when the user wants to supervise each step manually
|
||||
|
||||
## Example: Off vs On
|
||||
### When off
|
||||
Typical pattern:
|
||||
1. The agent finishes one visible step
|
||||
2. The agent reports that step
|
||||
3. The flow may pause until the user nudges again
|
||||
|
||||
### When on
|
||||
Typical pattern:
|
||||
1. The agent finishes one visible step
|
||||
2. The agent continues into the next low-risk step in the same session
|
||||
3. The user sees fewer unnecessary pauses during multi-step work
|
||||
|
||||
## Troubleshooting
|
||||
### `/ac` says unknown or does nothing
|
||||
Possible causes:
|
||||
- The running OpenClaw installation does not include the command yet
|
||||
- The chat is not hitting the expected installation/runtime
|
||||
- The sender is not authorized for command use on that surface
|
||||
|
||||
Check:
|
||||
- `/ac status`
|
||||
- If needed, inspect the live install rather than a repo checkout
|
||||
|
||||
### It still pauses sometimes
|
||||
That can be normal. Auto-continue does not override:
|
||||
- approval gates
|
||||
- safety constraints
|
||||
- destructive action confirmation
|
||||
- user-choice decision points
|
||||
|
||||
### It worked in one chat but not another
|
||||
This is expected when the other chat/session has not enabled it yet. The setting is session-scoped.
|
||||
|
||||
## Verification
|
||||
After enabling, verify with:
|
||||
```text
|
||||
/ac status
|
||||
```
|
||||
Expected result should indicate the session is no longer default and is using followup behavior.
|
||||
|
||||
A practical verification flow:
|
||||
1. Run `/ac on`
|
||||
2. Run `/ac status`
|
||||
3. Give the agent a multi-step but low-risk task
|
||||
4. Confirm the agent continues without waiting after every intermediate step
|
||||
|
||||
If you need concrete implementation evidence, the verified live installation path was:
|
||||
- `/home/chchang/.nvm/versions/node/v22.22.0/lib/node_modules/openclaw/dist/commands-handlers.runtime-Akj_Dqoi.js`
|
||||
|
||||
## Operator Guidance
|
||||
Recommend enabling it when:
|
||||
- the user wants smooth multi-step execution
|
||||
- the user dislikes repeated nudges
|
||||
- the work is mostly deterministic and low-risk
|
||||
|
||||
Recommend leaving it off when:
|
||||
- the user wants step-by-step supervision
|
||||
- the task has many decision gates
|
||||
- the task contains destructive or approval-heavy actions
|
||||
|
||||
## One-Line Rule
|
||||
**`/ac on` turns on session-level auto-followup for the current chat; `/ac off` turns it back off.**
|
||||
Reference in New Issue
Block a user