| name | retrospective |
| description | Analyze recent session context archives to identify coding patterns, agent usage, TDD cycle stats, and common EVAL issues. Generates a summary report with data-driven improvement suggestions. |
Session Retrospective
Overview
Analyze accumulated PLAN/ACT/EVAL session data from context archives to surface coding habits, recurring patterns, and actionable improvement suggestions. Transforms passive session history into data-driven growth insights.
Core principle: Decisions improve when informed by patterns, not just memory. Review what actually happened, not what you think happened.
When to Use
- After completing a sprint or milestone
- During periodic team/personal retrospectives
- When noticing repeated issues across sessions
- Before planning process improvements
- When onboarding to understand team patterns
When NOT to Use
- Mid-session (wait until a natural checkpoint)
- With fewer than 3 archived sessions (insufficient data)
- For real-time debugging (use
systematic-debugging instead)
Prerequisites
- Context archive system enabled (#999)
- At least 3 archived sessions in
docs/codingbuddy/archive/
- MCP tools available:
get_context_history, search_context_archives
Process
Phase 1: Data Collection
Gather session archives and extract structured data.
-
Retrieve recent archives
- Call
get_context_history with appropriate limit (default: 20)
- If no archives exist, inform user and suggest running a few PLAN/ACT/EVAL sessions first
- Note the date range covered
-
Read archive contents
- Read each archived context document
- Extract from each session:
- Mode transitions (PLAN, ACT, EVAL, AUTO)
- Primary agent used
- Task description and title
- Decisions made
- Notes recorded
- Progress items (ACT mode)
- Findings and recommendations (EVAL mode)
- Status (completed, in_progress, blocked)
Phase 2: Pattern Analysis
Analyze collected data across five dimensions.
2a. Mode Usage Patterns
- Count sessions per mode (PLAN, ACT, EVAL, AUTO)
- Calculate EVAL adoption rate:
EVAL sessions / total sessions
- Identify sessions that skipped EVAL (potential quality gaps)
- Flag AUTO mode usage frequency
2b. Agent Utilization
- Rank agents by frequency of use
- Identify underutilized specialists (available but rarely used)
- Detect agent concentration (over-reliance on one agent)
- Note any specialist gaps for the project's tech stack
2c. TDD Cycle Indicators
Search archives for TDD-related patterns:
search_context_archives with keywords: "TDD", "test", "RED", "GREEN", "REFACTOR"
- Count sessions mentioning test-first vs test-after
- Identify sessions where tests were skipped or deferred
- Calculate approximate TDD adherence rate
2d. Recurring Issues
Search for repeated problems:
search_context_archives with keywords: "blocked", "failed", "error", "issue", "bug"
- Group similar issues by category (build, test, integration, deployment)
- Identify issues that appear in 2+ sessions (systemic problems)
- Note resolution patterns (same fix applied repeatedly?)
2e. Decision Patterns
Analyze decisions across sessions:
- Extract all recorded decisions
- Identify decisions that were later reversed or modified
- Spot recurring decision themes (architecture, tooling, process)
- Flag decisions made without EVAL validation
Phase 3: Report Generation
Generate a structured markdown report.
# Session Retrospective Report
> Period: {start_date} - {end_date}
> Sessions analyzed: {count}
> Generated: {current_date}
## Summary Statistics
| Metric | Value |
|--------|-------|
| Total sessions | {n} |
| PLAN sessions | {n} |
| ACT sessions | {n} |
| EVAL sessions | {n} |
| AUTO sessions | {n} |
| EVAL adoption rate | {n}% |
| Completion rate | {n}% |
## Top Agents
| Rank | Agent | Sessions | % |
|------|-------|----------|---|
| 1 | {agent} | {n} | {n}% |
| ... | ... | ... | ... |
## TDD Health
- Adherence rate: {n}%
- Test-first sessions: {n}
- Test-after sessions: {n}
- No-test sessions: {n}
## Recurring Issues
### {Issue Category}
- **Frequency**: {n} sessions
- **Pattern**: {description}
- **Impact**: {assessment}
## Key Decisions Timeline
| Date | Decision | Mode | Validated? |
|------|----------|------|------------|
| {date} | {decision} | {mode} | {yes/no} |
## Improvement Suggestions
### High Priority
1. {suggestion with rationale}
### Medium Priority
1. {suggestion with rationale}
### Process Observations
1. {observation}
Phase 4: Improvement Suggestions
Generate actionable suggestions based on findings.
Auto-generate suggestions for:
| Pattern Detected | Suggestion |
|---|
| EVAL rate < 30% | Increase EVAL usage for quality gates |
| Single agent > 60% | Diversify specialist agents for broader coverage |
| TDD adherence < 50% | Reinforce test-first discipline with TDD skill |
| Same issue 3+ times | Create a checklist or rule to prevent recurrence |
| Blocked sessions > 20% | Investigate common blockers and add preventive steps |
| No decisions recorded | Improve decision documentation in context |
| Decisions reversed > 2x | Add EVAL validation before major decisions |
Phase 5: Output
- Display report in conversation
- Save report to
docs/codingbuddy/retrospective-{YYYY-MM-DD}.md
- Offer next steps:
- "Create GitHub issues for high-priority improvements?"
- "Update project rules based on findings?"
- "Schedule recurring retrospectives?"
Key Principles
- Data over opinion - Base all observations on archive evidence
- Patterns over incidents - Focus on recurring themes, not one-off events
- Actionable suggestions - Every finding should have a clear next step
- No blame - Focus on process improvement, not individual mistakes
- Incremental - Suggest 2-3 improvements per retrospective, not a complete overhaul