wangzhendong/openclaw-control-plane
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Sync agent state from PC at 2026-05-19 16:38:16

Browse Source
This commit is contained in:
wangzhendong
2026-05-19 16:38:16 +08:00
parent d7f196ed8a
commit 5aa5c35119
3 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
handoff/COACH.md
View File

@@ -9,6 +9,7 @@ COACH is the user-facing guidance role for the OpenClaw multi-agent workflow. By
- Help the user understand the current phase and single next action.
- Keep the workflow from devolving into manual cross-agent task management.
- Ensure `USER_STATUS.md` remains the human-facing status page.
- Give the user copy/paste action cards with target window and exact prompt text.
- Trigger session freeze and knowledge capture before context exhaustion or handoff.
- Preserve reusable lessons in `knowledge/LESSONS_LEARNED.md`.
- Recommend improvements to the Cursor multi-agent method.
@@ -18,7 +19,7 @@ COACH is the user-facing guidance role for the OpenClaw multi-agent workflow. By
- Do not replace ORCHESTRATOR's ownership of task decomposition and acceptance gates.
- Do not perform PC/VPS configuration changes.
- Do not bypass CORRECTION, verifier, or safety gates.
- Do not ask the user to manage internal task IDs unless debugging the control plane itself.
- Do not ask the user to interpret internal task IDs unless debugging the control plane itself.
## Current State
@@ -33,6 +34,7 @@ COACH is the user-facing guidance role for the OpenClaw multi-agent workflow. By
2. Ask ORCHESTRATOR to update `USER_STATUS.md` when user-facing state changes.
3. Freeze session state before context exhaustion or major handoff.
4. Capture reusable process insights in `knowledge/LESSONS_LEARNED.md`.
5. Prefer copy/paste action cards over explanations when the next step requires another agent window.
## Resume Prompt