/aggregate-logs

name: aggregate-logs
description: Generate LEARNINGS.md from skill execution logs over a configurable time window.

Aggregate Skill Execution Logs

Generate LEARNINGS.md from skill execution logs.

Usage

/abstract:aggregate-logs                    # Last 30 days (default)
/abstract:aggregate-logs --days 7           # Last 7 days
/abstract:aggregate-logs --days 90          # Last 90 days

Purpose

Part of Issue #69 Phase 3, this command processes skill execution logs to generate actionable insights for skill improvement.

What It Does

1. **Loads logs** from `~/.claude/skills/logs/<plugin>/<skill>/` 2. **Calculates metrics** for each skill:

• Execution count, success rate
• Average/max duration
• Average user rating (from evaluations)
• Common friction points
• Improvement suggestions

3. **Detects patterns**:

• High failure rates (>30%)
• Slow execution (>10s average)
• Low ratings (<3.5/5.0)

4. **Generates LEARNINGS.md** at `~/.claude/skills/LEARNINGS.md`

Output Format

LEARNINGS.md Structure

# Skill Performance Learnings

**Last Updated**: 2026-01-08 04:30:00 UTC
**Analysis Period**: Last 30 days
**Skills Analyzed**: 15
**Total Executions**: 342

## High-Impact Issues

### imbue:proof-of-work
**Type**: high_failure_rate
**Severity**: high
**Metric**: 42.3% success rate
**Detail**: 11/26 failures
**Recent Errors**:
- ValidationError: Missing acceptance_criteria field
- FileNotFoundError: PROOF.md not found
- KeyError: 'evidence' in evaluation

## Slow Execution

| Skill | Avg Duration | Max Duration | Executions |
|-------|--------------|--------------|------------|
| `sanctum:pr-agent` | 45.2s | 120.5s | 18 |
| `pensive:code-reviewer` | 32.1s | 89.3s | 24 |

## Low User Ratings

### abstract:skill-auditor - 2.8/5.0
**Common Friction**:
- Too verbose output
- Missing examples for modular skills
- Unclear token optimization guidance

**Improvement Suggestions**:
- Add --quiet flag
- Include example audits
- Link to token optimization docs

## Skill Performance Summary

| Skill | Executions | Success Rate | Avg Duration | Rating |
|-------|------------|--------------|--------------|--------|
| `imbue:proof-of-work` | 26 | 42.3% | 2.1s | 3.2/5.0 |
| `sanctum:pr-agent` | 18 | 94.4% | 45.2s | 4.5/5.0 |

Examples

Weekly Rollup

# Run every Monday morning
/abstract:aggregate-logs --days 7

# Output:
# Aggregating logs from last 7 days...
#
# ✅ LEARNINGS.md generated: ~/.claude/skills/LEARNINGS.md
#
# Summary:
#   Skills Analyzed: 8
#   Total Executions: 47
#   High-Impact Issues: 2
#   Slow Skills: 1
#   Low-Rated Skills: 1

Full History Analysis

/abstract:aggregate-logs --days 90

# Analyzes last 3 months of data

Integration

**Phase 2** (Evaluation):

• Reads `qualitative_evaluation` field from logs
• Calculates average ratings
• Aggregates friction points and suggestions

**Phase 4** (/fix-workflow):

• `/fix-workflow` reads LEARNINGS.md
• Surfaces known issues before manual analysis
• Prioritizes improvements by frequency × impact

**Phase 5** (/improve-skills):

• Uses LEARNINGS.md to identify improvement opportunities
• Prioritizes by severity (high-impact issues first)
• Tracks improvements over time (version comparison)

**Phase 6a** (Collective Intelligence):

• After LEARNINGS.md is generated, posts a summary to the target repo's Discussions
• Target repo is detected at runtime: a `target_repo` override in `~/.claude/skills/discussions/config.json`, otherwise the current repo from `gh repo view`
• Check opt-out: reads `~/.claude/skills/discussions/config.json`
• If `auto_post_learnings` is `true` (default), runs `post_learnings_to_discussions.py`
• Reports: "Posted learning summary to Discussions: {url}"
• Skips silently if `gh` is not authenticated or network is unavailable

Automation

Cron Job (Optional)

# Add to crontab for weekly rollup
0 9 * * MON cd /path/to/claude-night-market && ./plugins/abstract/scripts/aggregate_skill_logs.py 7

Manual Workflow

# After evaluating several skills:
/abstract:evaluate-skill proof-of-work
/abstract:evaluate-skill code-reviewer
/abstract:aggregate-logs

# Review LEARNINGS.md
cat ~/.claude/skills/LEARNINGS.md

# Act on insights
/abstract:improve-skills  # Phase 5

Metrics Thresholds

| Metric | Warning | Critical | |--------|---------|----------| | Success Rate | <80% | <70% | | Avg Duration | >5s | >10s | | User Rating | <4.0 | <3.5 | | Failure Count | >5 | >10 |

Performance

• **Runtime**: ~1-2s for 30 days of data (hundreds of executions)
• **Memory**: Loads one skill's logs at a time (streaming)
• **Disk**: LEARNINGS.md typically <50KB

Related

• `/abstract:evaluate-skill` - Capture qualitative feedback (Phase 2)
• `/abstract:improve-skills` - Act on insights (Phase 5)
• `/abstract:promote-discussions` - Promote highly-voted learnings to Issues (Phase 6c)
• `plugins/abstract/hooks/skill_execution_logger.py` -

Raw data capture (Phase 1)

Version

1.1.0 (Phase 6a integration)