| name | subagent-teams |
| description | How to run coordinated sub-agent teams (worker → reviewer → re-reviewer) and the Bedrock model-dispatch rules for sub-agents. Use this skill ONLY when spawning/dispatching sub-agents (e.g. Pi's Agent tool, parallel task fan-out) or debugging a sub-agent that dies instantly. Triggers on: "sub-agent", "subagent", "dispatch agents", "parallel agents", "Agent tool", "sub-agent finished in <2s / 0 tool uses".
|
Sub-Agent Teams
Use teams of coordinated sub-agents when possible to parallelize work and avoid
single-agent bottlenecks. Follow the three-step verdict pattern:
- Worker agent implements the change
- Reviewer agent analyzes architecture and finds issues
- Re-reviewer verifies the fix — catches lead-level errors
Sub-Agent Model Selection (Bedrock specifics)
When calling the Agent tool, omit the model parameter unless you must
override it — inherited-model sub-agents are guaranteed dispatchable (the
parent already proved it). Bedrock model IDs come in two forms:
- Bare ID (
anthropic.claude-...) needs provisioned throughput; on-demand
calls fail ("on-demand throughput isn't supported") and the sub-agent dies in
~1s with "completed: 0 tool uses".
- Cross-region inference profile (
us./eu./apac. prefix) supports
on-demand — this is what enabledModels in pi's settings should contain.
If you must pass model:, use a full us./eu./apac. profile ID, never a
bare anthropic.*. Fuzzy names (sonnet/haiku/opus) usually work but can resolve
to a bare ID; pi-extensions/safety-hooks.ts warns on fuzzy and blocks bare IDs.
A sub-agent finishing in <2s with 0 tool uses = model-dispatch failure —
verify the ID, drop the model: override, re-dispatch.