external-tool-coordination
npx machina-cli add skill a5c-ai/babysitter/external-tool-coordination --openclawExternal Tool Coordination
Overview
When configured via .metaswarm/external-tools.yaml, coordinate external AI tools for delegated implementation and cross-model adversarial review.
When to Use
- When external AI tools are configured and available
- For cross-model adversarial review (writer reviewed by different model)
- When delegating implementation to cheaper external models
- For escalation chains across tools
Capabilities
- Delegated Implementation - Send work units to external AI tools for implementation
- Cross-Model Review - Implementation by Model A, adversarial review by Model B
- Escalation Chains - Tool A -> Tool B -> Claude -> human
- Cost Optimization - Route simpler tasks to cheaper models
Configuration
Requires .metaswarm/external-tools.yaml:
tools:
- name: codex
type: openai
capabilities: [implement, review]
- name: gemini
type: google
capabilities: [implement, review]
escalation:
chain: [codex, gemini, claude, human]
Agents Used
agents/swarm-coordinator/- Multi-tool orchestration
Tool Use
Invoke as part of: methodologies/metaswarm/metaswarm-swarm-coordinator
Source
git clone https://github.com/a5c-ai/babysitter/blob/main/plugins/babysitter/skills/babysit/process/methodologies/metaswarm/skills/external-tool-coordination/SKILL.mdView on GitHub Overview
External Tool Coordination enables coordinating external AI tools like Codex and Gemini for delegated implementation and cross-model adversarial review when configured via .metaswarm/external-tools.yaml. It supports escalation chains and cost optimization by routing tasks to cheaper models.
How This Skill Works
The coordinator reads tools from external-tools.yaml, assigns work units to tools with implement and/or review capabilities, and manages cross-model reviews and escalations (codex -> gemini -> claude -> human). It relies on the swarm-coordinator agent for multi-tool orchestration.
When to Use It
- External tools are configured and reachable via .metaswarm/external-tools.yaml
- You need cross-model adversarial review where one model reviews another
- You want to delegate implementation work to cheaper external models
- You require escalation chains across tools when a task is high-risk
- You want to optimize costs by routing simpler tasks to cheaper models
Quick Start
- Step 1: Create .metaswarm/external-tools.yaml with tools and escalation chain
- Step 2: Invoke as part of: methodologies/metaswarm/metaswarm-swarm-coordinator
- Step 3: Monitor results and adjust tool assignments or escalation as needed
Best Practices
- Explicitly define tools and the escalation chain in external-tools.yaml
- Ensure each tool has compatible capabilities (implement, review) for the task
- Monitor latency and cost, and set timeouts and retry policies
- Validate and sanitize inputs/outputs before handing data between tools
- Use staged handoffs with clear success criteria and exit points
Example Use Cases
- Codex implements a feature; Gemini performs adversarial review on the output
- Writer output is generated by Codex and reviewed by Gemini for quality and safety
- Escalation: Codex -> Gemini -> Claude -> human on a risky task
- Cost optimization by routing simple tasks to Codex or Gemini, reserving Claude for complex work
- Multi-tool orchestration via agents/swarm-coordinator during a metaswarm sprint
