ccb-diagnose

Diagnose and recover stuck CCB agents using runtime authority and read-only pane inspection.

3.5k|342|Updated Oct 25, 2025
One-click install
npx skills add https://github.com/bfly123/claude_code_bridge --skill ccb-diagnose
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ccb-diagnose
Source: https://github.com/bfly123/claude_code_bridge/tree/main/inherit_skills/kimi_skills/ccb-diagnose
Command: npx skills add https://github.com/bfly123/claude_code_bridge --skill ccb-diagnose

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

When a CCB agent becomes stuck, disconnected, stops replying, or shows provider errors, this Skill provides a single structured workflow to diagnose the root cause, apply bounded recovery, verify the fix, and optionally file a redacted GitHub issue.

Core Features & Use Cases

  • Authority-Based Diagnosis: Combines ccb ping, ccb ps, ccb queue, ccb pend, and ccb trace output to establish the true runtime and job lineage state before acting.
  • Deep Pane Inspection: Classifies the agent's tmux pane state (working, waiting_input, stale_prompt, provider_error, dead_or_blank, and more) using read-only capture, never mutating panes.
  • Bounded Recovery: Applies the least disruptive supported repair such as ccb repair ack, ccb repair retry, ccb repair resubmit, or a guarded ccb restart, then verifies the result.
  • Use Case: A user reports that agent 'gemini' stopped replying mid-task. The Skill checks the daemon graph, inspects the pane, finds a provider rate-limit error, restarts the agent after cancelling stale work, verifies recovery, and prepares a redacted incident bundle.

Quick Start

Ask the assistant to run ccb-diagnose on the agent named gemini because it appears stuck and is not replying.

Frequently Asked Questions about ccb-diagnose

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagnose a stuck CCB agent?

Run ccb-diagnose with the agent name, for example $ccb_diagnose gemini. The workflow checks daemon authority with ccb ping, ps, queue, and trace, then inspects the agent's pane read-only to classify whether it is working, waiting for input, stale, or showing a provider error.

How do I fix a CCB agent that is not replying?

The Skill first establishes job lineage with ccb trace and queue details, then applies the least disruptive repair: ccb repair ack for broken acknowledgements, repair retry for incomplete attempts, repair resubmit for stale lineage, or a guarded ccb restart for dead provider panes.

Can ccb-diagnose restart all agents at once?

No. The Skill diagnoses exactly one named agent from the current daemon graph and never restarts all agents or runs project shutdown. If the agent name is missing or ambiguous, it asks for a specific current agent instead of inferring an all target.

Does ccb-diagnose submit GitHub issues automatically?

No. It creates a local redacted incident bundle, shows the proposed issue title and body, and only submits via gh issue create after explicit user confirmation of that exact redacted content. Credentials, full prompts, and pane dumps are never included.

Why does ccb-diagnose refuse to use tmux send-keys?

Pane mutation is a red line because pane text is evidence, not authority. The Skill only uses read-only capture through CCB-owned pane tools or a guarded tmux capture-pane fallback, preventing accidental disruption of running provider sessions.