diagnose-why-work-stopped

Diagnose stalled or looping agent issue trees and produce approval-gated remediation plans.

Updated Jul 14, 2026
One-click install
npx skills add https://github.com/Douglas-v/strategic-ai-paperclip-lab --skill diagnose-why-work-stopped-douglas-v
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: diagnose-why-work-stopped
Source: https://github.com/Douglas-v/strategic-ai-paperclip-lab/tree/main/paperclip/skills/diagnose-why-work-stopped
Command: npx skills add https://github.com/Douglas-v/strategic-ai-paperclip-lab --skill diagnose-why-work-stopped-douglas-v

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When an agent-orchestrated issue tree stalls, loops infinitely, or recovers too deeply, teams need a repeatable forensic procedure to find the exact stop point and design a general product rule — without shipping premature code changes. ## Core Features & Use Cases - Forensic Tree Walk: Walks the named issue tree node-by-node to identify the exact issue and state combination that stopped progress, quoting run ids and comment timestamps as evidence. - Invariant-Checked Rule Design: Frames every fix as a general product contract that preserves three invariants: productive work continues, only real blockers stop work, and no infinite loops. - Approval-Gated Planning: Writes a phased plan into the issue's plan document, opens a request_confirmation interaction, and only creates child issues after board/CTO acceptance. - Use Case: A manager links a stalled issue tree and asks why it stopped. The skill produces a root-cause write-up classifying every non-progressing issue as human-needed, agent-actionable, or already-covered, plus a phased remediation plan awaiting approval. ## Quick Start Ask the agent to diagnose why the linked issue tree stopped working and produce a root-cause analysis with a phased fix plan for board approval.

Frequently Asked Questions about diagnose-why-work-stopped

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

FAQPage Schema
How do I diagnose why an agent issue tree stopped working?

Walk the tree node-by-node to find the exact issue and state combination that stops progress, quoting run ids and comment timestamps as evidence. Common stop shapes include in_review states with no participant, cancelled blocker leaves, and unbounded continuation recovery.

How to prevent infinite loops in agent work recovery?

Bound continuation and stranded-work recovery loops, and ensure recovery never treats its own recovery issues as recoverable source work. Every proposed rule must be checked against the no-infinite-loops invariant before being included in the plan.

When should I use forensic diagnosis instead of fixing the bug directly?

Use this procedure when the assignment asks why work stopped, why it loops, or requests a general product rule before any change. If the task is a direct code fix or a normal bug report on a specific feature, use the standard engineering flow instead.

Can this procedure make code changes to fix the stalled tree?

No. The output is a written root cause and a phased plan gated by a request_confirmation interaction. Child issues and code changes are created only after board or CTO acceptance of the plan.

What are the three invariants for agent liveness rules?

Productive work must continue without user intervention, only real blockers may stop work, and recovery loops must be bounded. Every diagnosis and proposed rule must explicitly show how all three invariants are preserved together.