Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From quorum
Handles correction cycles for rejected agent tasks: sends feedback via SendMessage to original agent for lint/test fixes. Escalates security/repeated failures. Manages upstream delays and blocks downstream tasks.
npx claudepluginhub berrzebb/quorum --plugin quorumHow this agent operates — its isolation, permissions, and tool access model
Agent reference
quorum:agents/knowledge/references/orchestrator/correctionThe summary Claude sees when deciding whether to delegate to this agent
On `[pending_tag]` rejection — **send correction to existing agent via SendMessage**. 1. Look up `agent_id` for the task in handoff 2. Read rejection codes + rationale from respond file 3. Compose correction prompt: ``` SendMessage(to: "<agent_id>") { } ``` 4. Update handoff status: `auditing` → `correcting` 5. Agent fixes and resubmits → re-enters audit loop | Rejection Type | Action | |------...Fills Nyquist validation gaps by generating runnable behavioral tests for phase requirements, running them adversarially, debugging failures (max 3 iterations), verifying coverage, and escalating blockers.
Share bugs, ideas, or general feedback.
On [pending_tag] rejection — send correction to existing agent via SendMessage.
agent_id for the task in handoffSendMessage(to: "<agent_id>") {
## Correction Round: [task-id]
### Rejection Codes: ...
### Instructions: ...
}
auditing → correcting| Rejection Type | Action |
|---|---|
| CQ (lint/type) | SendMessage — same agent, minor fix |
| T (test failure) | SendMessage — same agent |
| CC (mismatch) | SendMessage — same agent, rewrite evidence |
| security/regression | Escalate to user — high risk |
| 3+ repeated rejections | Escalate to user — approach needs rethinking |
If the agent has terminated or is unresponsive:
agent_id in handoffWhen parallel tracks are active, monitor for upstream delays:
audit_history for the current trackblocked (upstream: [task-id] rejected 3x)delayedIn headless mode, the orchestrator cannot ask the user "escalate or continue?" Instead:
blocked (needs human review) + stop spawning