A bug triage skill: four investigators and an evidence check

---
name: bug-triaging-graph
description: Triages a bug report with four parallel investigators, re-checks the evidence in a fresh context, and writes a root-cause diagnosis with a test plan.
spec: skill-graph/1
topology:
  nodes:
    - id: goal
      type: input
      label: BUG REPORT
    - id: splitter
      type: orchestrator
      label: SPLIT
      model: sonnet
      tools:
        - Read
    - id: stack_trace
      type: worker
      label: Stack Trace Analyzer
      model: haiku
      tools:
        - Read
        - Grep
        - Glob
        - Bash
    - id: logs_metrics
      type: worker
      label: Log / Metric Correlator
      model: haiku
      tools:
        - Read
        - Grep
        - Glob
        - Bash
    - id: commits
      type: worker
      label: Commit Inspector
      model: haiku
      tools:
        - Read
        - Grep
        - Glob
        - Bash
    - id: repro
      type: worker
      label: Minimal Repro Builder
      model: haiku
      tools:
        - Read
        - Grep
        - Glob
        - Bash
    - id: verifier
      type: verifier
      label: VERIFIER (Fresh Context)
      isolated_context: true
      model: opus
      tools:
        - Read
        - Grep
        - Glob
      gate: At least one hypothesis is connected to the symptom by evidence, and the repro reproduces it.
      on_fail: retry
      max_retries: 1
    - id: merger
      type: synthesizer
      label: FINAL ANSWER
      model: sonnet
      tools:
        - Read
  edges:
    - from: goal
      to: splitter
    - from: splitter
      to:
        - stack_trace
        - logs_metrics
        - commits
        - repro
      contract: subtask_json
    - from:
        - stack_trace
        - logs_metrics
        - commits
        - repro
      to: verifier
      contract: hypotheses
    - from: verifier
      to: merger
      contract: root_cause
contracts:
  subtask_json:
    format: json
    description: One key per worker id; each value is that worker's task.
  hypotheses:
    format: markdown
    description: A numbered list; every item has its evidence and a confidence of high, medium or low.
  root_cause:
    format: markdown
    description: Only the items that passed verification, then the rejected ones with the reason.
---

# Node Prompts & Edge Contracts

## @node: splitter
You are the triage lead. Split the bug report into four investigation tasks.

Input:
{{goal}}

Output strictly as JSON with one key per worker: `stack_trace`, `logs_metrics`, `commits`, `repro`. Each value is the exact task for that worker.

## @node: stack_trace
Analyze ONLY the stack trace and error: the failing frame, the inputs that reach it, and why it throws.

Your task: {{splitter.stack_trace}}

Return findings as a numbered list. For each one give the evidence (file, line, quote or source) and a confidence of high, medium or low.

## @node: logs_metrics
Correlate ONLY logs and metrics: when it started, what changed at that time, and which users or requests are affected.

Your task: {{splitter.logs_metrics}}

Return findings as a numbered list. For each one give the evidence (file, line, quote or source) and a confidence of high, medium or low.

## @node: commits
Inspect ONLY recent commits and config changes that touch the failing path, and rank them by likelihood.

Your task: {{splitter.commits}}

Return findings as a numbered list. For each one give the evidence (file, line, quote or source) and a confidence of high, medium or low.

## @node: repro
Build ONLY a minimal reproduction: the smallest steps or script that trigger the bug.

Your task: {{splitter.repro}}

Return findings as a numbered list. For each one give the evidence (file, line, quote or source) and a confidence of high, medium or low.

## @node: verifier
You are an independent debugger. Check that the evidence actually connects each hypothesis to the symptom; reject the rest.

You only see the raw worker outputs below, not their reasoning. Drop any claim without concrete evidence, merge duplicates, and mark what you rejected and why.

Original input: {{goal}}

Worker outputs:
{{workers.output}}

## @node: merger
Write the diagnosis: root cause, evidence, fix, and a test plan that reproduces the bug before the fix and passes after it.

{{verifier.output}}