By name: d-verify description: Verify hypotheses with debug logging. Updates ./.gtd/debug/current/ROOT_CAUSE.md disable-model-invocation: true
Core responsibilities:
- Load hypotheses in confidence order
- Add strategic debug logs to test each hypothesis
- Run reproduction steps
- Analyze debug output
- Move to next hypothesis if rejected
- Document root cause when found
Flow: Load Hypotheses → Test Highest Confidence → Analyze → Found or Next
./.gtd/debug/current/SYMPTOM.md— Reproduction steps./.gtd/debug/current/HYPOTHESES.md— Hypotheses to test
Output:
./.gtd/debug/current/ROOT_CAUSE.md— When found- Debug logs in code (temporary)
One Hypothesis at a Time
Test systematically. Don't add logs for all hypotheses at once.
Strategic Logging
Add logs that can definitively confirm or reject the hypothesis.
Evidence-Based Conclusion
Root cause must be backed by debug output, not assumption.
Know When to Stop
If all hypotheses are rejected, stop and ask user to inspect again.
1. Load Context
Read both files:
./.gtd/debug/current/SYMPTOM.md— For reproduction steps./.gtd/debug/current/HYPOTHESES.md— For hypotheses list
if ! test -f "./.gtd/debug/current/SYMPTOM.md" || ! test -f "./.gtd/debug/current/HYPOTHESES.md"; then
echo "Error: Missing required files"
exit 1
fi
2. Spawn Verification Agents
For each hypothesis, spawn a researcher to inspect code for confirmation/rejection (READ-ONLY).
Concurrency: As many as needed (Parallel). Trigger: After loading hypotheses.
Fill prompt for each hypothesis:
<objective>
Verify Hypothesis: {hypothesis_description}
</objective>
<context>
- Symptom: {symptom_description}
- Evidence: {existing_evidence}
</context>
<investigation_checklist>
1. Locate relevant code
2. Check for logic errors matching hypothesis
3. Verify if preconditions/assumptions hold true
4. Identify where to add debug logs if needed
</investigation_checklist>
<output_format>
Verification Report:
- Status: Confirmed / Rejected / Inconclusive
- Evidence Found: (Code snippets, logic analysis)
- Recommended Debug Log Location: (File:Line)
</output_format>
Task(
prompt=filled_prompt,
subagent_type="researcher",
description="Verifying hypothesis: {short_desc}"
)
3. Review & Decide Actions
Review reports from all subagents.
3a. If Root Cause Confirmed by Inspection
If a subagent found definitive proof (e.g., obvious logic bug):
- Skip logging.
- Proceed to Document Root Cause.
3b. If Inconclusive / Needs Logs
Select the most promising hypothesis based on subagent findings.
Action:
- Add debug logs to locations recommended by subagent.
- Run reproduction steps.
- Analyze output.
3c. If All Rejected
If all subagents report "Rejected" with strong evidence:
- STOP.
- Ask user to re-inspect or provide new ideas.
4. Document Root Cause
When confirmed (by inspection or logs), write to ./.gtd/debug/current/ROOT_CAUSE.md:
# Root Cause
**Found:** {date}
**Status:** CONFIRMED
## Root Cause
{Clear description of the actual root cause}
## Verified Hypothesis
**Original Hypothesis {N}:** {description}
**Confidence:** {original percentage} → **Confirmed**
## Evidence
{Debug output and observations that confirmed this}
**Debug logs showed:**
- {key finding 1}
- {key finding 2}
## Location
- **Files:** `{file1}`, `{file2}`
- **Lines:** {line ranges}
- **Function/Method:** {specific location}
## Why It Causes The Symptom
{Explain the causal chain from root cause to observed symptom}
## Rejected Hypotheses
{List other hypotheses tested and why they were rejected}
4a. Clean Up Debug Logs
Remove or comment out temporary debug logs added during verification.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GTD:DEBUG ► ROOT CAUSE FOUND ✓
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Root cause documented: ./.gtd/debug/current/ROOT_CAUSE.md
Verified hypothesis: {N}
Location: {files}
─────────────────────────────────────────────────────
▶ Next Up
/d-plan-fix — create fix plan
─────────────────────────────────────────────────────