exporting-ai-reports

name	exporting-ai-reports
version	v1.1.0
description	Automated 3-AI verification result documentation and export workflow. Triggers when user requests AI verification report export, documentation of findings, or saving cross-validation results. Use after completing Codex/Gemini/Qwen analysis.

AI Verification Report Export

Target Token Efficiency: 78% (450 tokens → 99 tokens)

Purpose

Automated 3-AI verification result documentation without manual formatting or file organization.

Trigger Keywords

• "export AI report"
• "document findings"
• "save verification results"
• "AI 검증 결과"
• "3-AI 결과"

Context

• Project: OpenManager VIBE v5.80.0
• AI Tools: Codex, Gemini, Qwen (3-AI cross-verification)
• Output Location: logs/ai-decisions/
• Average Decision Quality: 9.2/10 (based on validation history)

Workflow

1. Identify AI Outputs

Required Information:

• Codex analysis (실무 검증)
• Gemini review (아키텍처 검증)
• Qwen assessment (성능 검증)
• Task/feature being verified
• Verification timestamp

Sources:

/tmp/codex.txt (from codex-wrapper.sh)
/tmp/gemini.txt (from gemini-wrapper.sh)
/tmp/qwen.txt (from qwen-wrapper.sh)

2. Parse AI Responses

Extract Key Data:

• Each AI's score (x/10)
• Key findings (3-5 points each)
• Consensus points (agreements)
• Divergent points (disagreements)
• Recommended actions

Automated Parsing (Enhancement 1):

# Parse scores from AI outputs
CODEX_SCORE=$(awk '/score:|점수:/ {print $NF}' /tmp/codex.txt 2>/dev/null | grep -oE '[0-9]+\.[0-9]+|[0-9]+' | head -1)
GEMINI_SCORE=$(awk '/score:|점수:/ {print $NF}' /tmp/gemini.txt 2>/dev/null | grep -oE '[0-9]+\.[0-9]+|[0-9]+' | head -1)
QWEN_SCORE=$(awk '/score:|점수:/ {print $NF}' /tmp/qwen.txt 2>/dev/null | grep -oE '[0-9]+\.[0-9]+|[0-9]+' | head -1)

# Calculate average score
if [ -n "$CODEX_SCORE" ] && [ -n "$GEMINI_SCORE" ] && [ -n "$QWEN_SCORE" ]; then
  AVERAGE_SCORE=$(echo "scale=1; ($CODEX_SCORE + $GEMINI_SCORE + $QWEN_SCORE) / 3" | bc 2>/dev/null || echo "0")
else
  AVERAGE_SCORE="N/A"
  echo "⚠️  WARNING: Unable to parse all AI scores"
fi

# Extract key findings (first 5 bullet points from each AI)
CODEX_FINDINGS=$(grep -E "^- |^\* " /tmp/codex.txt 2>/dev/null | head -5)
GEMINI_FINDINGS=$(grep -E "^- |^\* " /tmp/gemini.txt 2>/dev/null | head -5)
QWEN_FINDINGS=$(grep -E "^- |^\* " /tmp/qwen.txt 2>/dev/null | head -5)

Consensus Detection (Enhancement 2):

# Extract common keywords across all 3 AI outputs
# Simple approach: Find keywords appearing in at least 2 outputs
CODEX_KEYWORDS=$(tr '[:upper:]' '[:lower:]' < /tmp/codex.txt | grep -oE '[a-z]{4,}' | sort | uniq)
GEMINI_KEYWORDS=$(tr '[:upper:]' '[:lower:]' < /tmp/gemini.txt | grep -oE '[a-z]{4,}' | sort | uniq)
QWEN_KEYWORDS=$(tr '[:upper:]' '[:lower:]' < /tmp/qwen.txt | grep -oE '[a-z]{4,}' | sort | uniq)

# Find intersection (keywords in at least 2 AIs)
CONSENSUS_KEYWORDS=$(echo "$CODEX_KEYWORDS $GEMINI_KEYWORDS $QWEN_KEYWORDS" | tr ' ' '
' | sort | uniq -c | awk '$1 >= 2 {print $2}')

# Generate consensus summary
CONSENSUS_SUMMARY="Common themes: $(echo $CONSENSUS_KEYWORDS | head -10 | tr '
' ', ' | sed 's/, $//')"

Status Determination (Enhancement 3):

# Threshold-based approval logic
if [ "$AVERAGE_SCORE" != "N/A" ]; then
  if (( $(echo "$AVERAGE_SCORE >= 9.0" | bc -l) )); then
    STATUS="✅ APPROVED"
    STATUS_REASON="High consensus (≥9.0), implementation recommended"
  elif (( $(echo "$AVERAGE_SCORE >= 8.0" | bc -l) )); then
    STATUS="⚠️  CONDITIONALLY APPROVED"
    STATUS_REASON="Good score (≥8.0), minor improvements suggested"
  elif (( $(echo "$AVERAGE_SCORE >= 7.0" | bc -l) )); then
    STATUS="🔄 NEEDS REVISION"
    STATUS_REASON="Moderate score (≥7.0), significant improvements required"
  else
    STATUS="❌ REJECTED"
    STATUS_REASON="Low score (<7.0), major redesign recommended"
  fi
else
  STATUS="⚠️  INCOMPLETE"
  STATUS_REASON="Unable to calculate average score (missing AI outputs)"
fi

Template Structure:

# [Task Name] - 3-AI Verification

**Date**: YYYY-MM-DD HH:mm KST
**Status**: [APPROVED / CONDITIONALLY APPROVED / REJECTED]

## Scores

- Codex (실무): X.X/10
- Gemini (아키텍처): X.X/10
- Qwen (성능): X.X/10
- **Average**: X.X/10

## Consensus

[Agreed points across all 3 AIs]

## Key Findings

### Codex (실무 검증)

- Finding 1
- Finding 2
- Finding 3

### Gemini (아키텍처 검증)

- Finding 1
- Finding 2
- Finding 3

### Qwen (성능 검증)

- Finding 1
- Finding 2
- Finding 3

## Divergent Views

[Points where AIs disagree, if any]

## Recommended Actions

1. Priority 1: [Action]
2. Priority 2: [Action]
3. Priority 3: [Action]

## Decision Rationale

[Claude's final decision based on 3-AI input]

3. Generate Report File

Filename Convention:

logs/ai-decisions/YYYY-MM-DD-{task-slug}.md

Example:

logs/ai-decisions/2025-11-04-skills-implementation-verification.md

✅ Enhancement 5: Filename Generation

# Auto-generate filename slug from task name
# Input: TASK_NAME="Skills Implementation Verification"
# Output: skills-implementation-verification
TASK_SLUG=$(echo "$TASK_NAME" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | sed 's/--*/-/g' | sed 's/^-//' | sed 's/-$//')

# Generate full filename with timestamp
FILENAME="logs/ai-decisions/$(date +%Y-%m-%d)-${TASK_SLUG}.md"

echo "📝 Report filename: $FILENAME"

Slug Generation Logic:

• Convert to lowercase: tr '[:upper:]' '[:lower:]'
• Replace non-alphanumeric with hyphens: sed 's/[^a-z0-9]/-/g'
• Collapse multiple hyphens: sed 's/--*/-/g'
• Trim leading hyphen: sed 's/^-//'
• Trim trailing hyphen: sed 's/-$//'

4. Export to Repository

✅ Enhancement 4: Validation Workflow

# Check file existence before parsing
MISSING=""
[ ! -f /tmp/codex.txt ] && MISSING="${MISSING}codex "
[ ! -f /tmp/gemini.txt ] && MISSING="${MISSING}gemini "
[ ! -f /tmp/qwen.txt ] && MISSING="${MISSING}qwen "

if [ -n "$MISSING" ]; then
  echo "⚠️  WARNING: Missing AI outputs: $MISSING"
  echo "   Report will be incomplete. Run wrapper scripts first:"
  echo "   ./scripts/ai-subagents/codex-wrapper.sh \"[query]\""
  echo "   ./scripts/ai-subagents/gemini-wrapper.sh \"[query]\""
  echo "   ./scripts/ai-subagents/qwen-wrapper.sh \"[query]\""
  exit 1
fi

# Validate AI output format (non-empty files)
for ai_file in /tmp/codex.txt /tmp/gemini.txt /tmp/qwen.txt; do
  if [ ! -s "$ai_file" ]; then
    echo "❌ ERROR: Empty AI output file: $ai_file"
    exit 1
  fi
done

echo "✅ Validation passed: All AI outputs present and non-empty"

File Location:

• Primary: logs/ai-decisions/
• Backup: /tmp/ai-verification-backup/

Verify Export:

• ✅ File created in correct directory
• ✅ Markdown formatting valid
• ✅ All 3 AI outputs included
• ✅ Timestamp accurate
• ✅ Filename follows convention

5. Report Summary

Format:

📝 AI Verification Report Exported

📊 Summary:
├─ Task: [Task Name]
├─ Average Score: X.X/10
├─ Status: [APPROVED / CONDITIONAL / REJECTED]
├─ File: logs/ai-decisions/YYYY-MM-DD-{task-slug}.md
└─ Duration: Xs

✅ Next Steps:
- Review consensus points
- Implement recommended actions
- Update task status

Token Optimization Strategy

Before (Manual):

User: "3-AI 검증 결과를 정리해줘"
Assistant: [reads /tmp/*.txt, formats markdown, asks for filename, creates file, explains structure]
Tokens: ~450

After (Skill):

User: "export AI report"
Skill: [parses outputs, auto-generates filename, creates file, reports summary]
Tokens: ~99 (78% reduction)

Efficiency Gains:

• ❌ No need to explain report structure
• ❌ No need to manually format markdown
• ✅ Auto-parse AI outputs
• ✅ Auto-generate filename
• ✅ Structured export process

Common Report Types

Type 1: Implementation Verification

Task: Feature X Implementation
Status: APPROVED (9.2/10)
Consensus: Architecture SOLID-compliant, performance acceptable
Actions: Minor refactoring suggestions

Type 2: Bug Analysis

Task: Bug Y Root Cause Analysis
Status: CONDITIONALLY APPROVED (8.5/10)
Consensus: Root cause identified, fix proposed
Divergence: Qwen suggests additional optimization
Actions: Implement fix + consider optimization

Type 3: Architecture Review

Task: System Z Design Review
Status: REJECTED (6.8/10)
Consensus: Major SOLID violations detected
Critical Issues: 5 violations, 3 security risks
Actions: Redesign required

Edge Cases

Case 1: Missing AI Output

• Check: Verify all 3 files exist (/tmp/codex.txt, /tmp/gemini.txt, /tmp/qwen.txt)
• Action: Report which AI outputs are missing
• Fallback: Export available outputs only, note incomplete verification

Case 2: Parsing Errors

• Check: Validate AI output format (score patterns, structured responses)
• Action: Report parsing issues, extract raw outputs
• Fallback: Manual review required flag in report

Case 3: Duplicate Reports

• Check: Verify filename doesn't already exist
• Action: Append timestamp suffix (-v2, -v3)
• Prevention: Unique task slugs

Case 4: Large Outputs

• Check: AI output file sizes > 50KB
• Action: Truncate to key findings only
• Include: Full output reference path

Success Criteria

• Report generated: < 2 min
• All 3 AI outputs included: 100%
• Markdown formatting valid: ✅
• Filename convention followed: ✅
• No manual intervention required

Related Skills

• tests/lint-smoke.md - If testing verification results
• performance/next-router-bottleneck.md - If performance analysis verification

Changelog

• 2025-11-04: Initial implementation (Phase 1)