dag-replay-debugger

What problem does it solve?

Post-mortem DAG runs can hide where a decision or reasoning branch went wrong because inputs, prompts, and intermediate state are hard to inspect and compare; this Skill exposes node-level state and enables selective replay so engineers can find the divergence point quickly.

Core Features & Use Cases

• State Inspection: View a node's inputs, system/user prompts, full output, evaluator scores, context store entries, timing, and cost for any completed execution.
• Checkpoint Replay: Replay a run from any completed node with options to keep inputs, edit upstream outputs, swap skills, or change model tiers to evaluate alternate outcomes.
• Execution Diffing: Compare original and replayed traces side-by-side to spot missing outputs, model drift, routing mistakes, or cost/quality tradeoffs.
• Reasoning Trace Analysis: Surface thinking tokens or chain-of-thought where available to reveal why a node made a particular decision.
• Use Case: Diagnose why a multi-node agent pipeline produced an incorrect final decision by inspecting node 7's state, editing its input, and replaying downstream nodes to validate the fix.

Quick Start

Inspect node analyze-codebase in a completed run, then choose Edit & Replay to modify the upstream output and compare the new trace to the original.