30 lines
1.1 KiB
Markdown
30 lines
1.1 KiB
Markdown
---
|
|
name: force-recall
|
|
description: "Prepend mandatory RULEBOOK/SOUL recall block before the agent sees inbound messages"
|
|
homepage: https://docs.openclaw.ai/automation/hooks
|
|
metadata:
|
|
{ "openclaw": { "emoji": "🧠", "events": ["message:preprocessed"], "always": true } }
|
|
---
|
|
|
|
# Force Recall Hook (MVP)
|
|
|
|
This hook enforces a **recall gate** by prepending a short, high-salience block to every inbound message *after* media/link enrichment and *before* the agent sees it.
|
|
|
|
Goal: **Before any technical action/tooling**, the agent must recall key rules from `docs/RULEBOOK.md` + `SOUL.md`.
|
|
|
|
## Behavior
|
|
|
|
- Listens on `message:preprocessed`
|
|
- Injects a `RECALL_GATE` prefix into `context.bodyForAgent`
|
|
- Optional debug: set `OPENCLAW_FORCE_RECALL_DEBUG=1` to append a one-line marker (visible in the agent prompt)
|
|
|
|
## Why this MVP
|
|
|
|
OpenClaw hooks currently provide reliable interception at the message boundary (`message:preprocessed`). This is the earliest stable point to force rules into the model's working context without patching core.
|
|
|
|
## Disable
|
|
|
|
```bash
|
|
openclaw hooks disable force-recall
|
|
```
|