session-review

# Session Review Skill

Analyze agent sessions to extract lessons and improve future performance.

## Purpose

After an agent session (or series of sessions), this skill helps you:
- Understand what the agent actually did
- Identify good patterns to reinforce
- Spot bad patterns to prevent
- Find wasted effort
- Generate improvements for skills/prompts

## Quick Start

### Trigger Phrases

- "Review the last session"
- "Analyze what went wrong in this project"
- "What patterns do you see in the agent's work?"
- "Help me understand why the agent struggled with X"

### Basic Usage

```
1. Run after completing a project or hitting issues
2. Point to project with .claude/ folder
3. Get structured analysis with recommendations
```

## Input Sources

The skill reads from multiple sources:

```
.claude/
├── progress.md           # Session summaries
├── sessions/             # Detailed session logs
│   ├── 2025-01-10-001.md
│   └── 2025-01-10-002.md
├── features.json         # Original plan
├── tasks.db              # Task tracking
└── checkpoints/          # Human decision points
    └── pending-review.md

Git History
├── Commit messages       # What was done
├── Diffs                 # What actually changed
└── Reverts               # What was undone

Code Quality
├── Build status          # Did it compile?
├── Test results          # Did tests pass?
└── Lint output           # Code quality issues
```

## Analysis Framework

### 1. Execution Analysis

Compare plan vs reality:

| Metric | Source | Question |
|--------|--------|----------|
| Tasks planned | features.json | How many tasks were defined? |
| Tasks completed | progress.md | How many actually finished? |
| Tasks abandoned | git history | What was started but reverted? |
| Scope creep | task additions | What wasn't in original plan? |
| Time estimates | tasks.db | How accurate were estimates? |

### 2. Quality Analysis

Assess work quality:

| Signal | Good | Bad |
|--------|------|-----|
| Build failures | 0-1 p