multi-agent-coordinator
The multi-agent-coordinator skill manages parallel OpenClaw agents by maintaining a shared registry, detecting timeouts through heartbeat monitoring, comparing outputs for consistency conflicts, and enabling structured context handoffs between agents. Use this when orchestrating three or more independent agents that need coordinated state tracking, contradiction detection, and managed information flow across separate channels and sessions.
git clone --depth 1 https://github.com/ArchieIndian/openclaw-superpowers /tmp/multi-agent-coordinator && cp -r /tmp/multi-agent-coordinator/skills/openclaw-native/multi-agent-coordinator ~/.claude/skills/multi-agent-coordinatorSKILL.md
# multi-agent-coordinator
When running 3+ OpenClaw agents in parallel, the flat `agents.list[]` config becomes unmanageable, channel "bleeding" occurs between agents, parallel agents can produce contradictory outputs, and there is no timeout detection for silent agent failures.
This skill lives in the **orchestrator agent** and provides: a shared agent registry, health-check heartbeats, output consistency checks, and structured inter-agent handoffs.
## Agent registry
Register each sub-agent when it starts:
```
python3 run.py --register agent-id=coder role=code_implementation channel=C001
python3 run.py --register agent-id=reviewer role=code_review channel=C002
```
The registry tracks: agent ID, role, channel, status, last-seen timestamp, and current task.
## Health checks
The orchestrator pings each registered agent's last-seen timestamp. An agent is considered **timed out** if it hasn't updated in longer than the configured timeout (default: 30 minutes).
Invoke periodically:
```
python3 run.py --health-check
```
On timeout detection: write `AGENT_TIMEOUT` to state, alert the orchestrator, and prompt: "Agent `coder` hasn't responded in 45 minutes. Reassign task or wait?"
## Output consistency checkpoint
Before merging parallel agent outputs:
```
python3 run.py --consistency-check --agents coder reviewer --key "api_design"
```
The orchestrator compares outputs for the same key and flags contradictions. Example: `coder` proposes `POST /users`, `reviewer` proposes `PUT /users/{id}` for the same operation — these are flagged as contradictions requiring human resolution before merge.
## Structured handoff
Pass context from one agent to another:
```
python3 run.py --handoff --from coder --to reviewer --task-file handoff.yaml
```
Handoff files capture: what was done, what wasn't, key decisions made, constraints discovered, and next recommended action. This extends `task-handoff` (which is single-agent) to the multi-agent case.
## Difference from workflow-orchestration
`workflow-orchestration` executes a linear sequence of steps in a single agent. `multi-agent-coordinator` manages a parallel fleet: independent agents with their own sessions, channels, and context windows.Syncs agent daily memory and MEMORY.md to an Obsidian vault so notes are human-browsable. Use nightly or on demand.
Structured ideation before any implementation. Use when starting any non-trivial task.
Scaffolds and validates new superpowers skills. Use when creating a new skill for this repository.
Executes plans task-by-task with verification. Use when implementing a plan.
Triggers a secondary verification pass for any agent output containing factual claims, numbers, dates, or named entities before the output is acted on
Crawls a new codebase to infer stack, conventions, and key invariants, then generates a PROJECT.md context file for the agent
Handles PR review feedback by fetching comments, grouping issues, fixing one group at a time, and verifying before replies.
Detects skill name shadowing and description-overlap conflicts that cause OpenClaw to trigger the wrong skill or silently ignore one when two skills compete for the same intent.