retro

# Retro Skill

Extract learnings, patterns, and insights from completed work.

## Overview

Transform session knowledge into durable, searchable artifacts:
- `~/gt/.agents/<rig>/retros/` - Session retrospective summaries
- `~/gt/.agents/<rig>/learnings/` - Reusable knowledge extracted
- `~/gt/.agents/<rig>/patterns/` - Reusable patterns discovered

**When to Use**:
- After completing significant work (epic closed)
- After encountering unexpected challenges
- After discovering useful patterns

---

## Phase 0: Rig Detection

**CRITICAL**: All `.agents/` artifacts go to `~/gt/.agents/<rig>/` based on the work being retrospected.

**Detection Logic**:
1. Identify which rig the completed work was in (from beads prefix or file paths)
2. If retro covers multiple rigs, use `_cross-rig`
3. If unknown/unclear, ask user

| Work Context | Target Rig | Output Base |
|--------------|------------|-------------|
| `ap-*` issues | `ai-platform` | `~/gt/.agents/ai-platform/` |
| `ka-*` issues | `kagent` | `~/gt/.agents/kagent/` |
| `gt-*` issues | `gastown` | `~/gt/.agents/gastown/` |
| Multiple rigs | `_cross-rig` | `~/gt/.agents/_cross-rig/` |

```bash
# Set RIG variable for use in output paths
RIG="ai-platform"  # or kagent, gastown, _cross-rig
mkdir -p ~/gt/.agents/$RIG/{retros,learnings,patterns}/
```

---

## Workflow

```
1. Gather Context      -> Analyze commits, beads, blackboard
1.5. Query Analytics   -> GET /memories/analytics/sources (NEW)
2. Identify Friction   -> Detect problems and patterns
3. Propose Changes     -> Generate concrete improvements
4. Supersession        -> Check for outdated artifacts
5. User Review         -> Present proposals for approval
6. Apply Changes       -> Execute approved edits
7-9. Write Outputs     -> Retro, learnings, patterns (include Source Performance)
10-11. Finalize        -> Update blackboard, confirm
```

---

## Phase 1-2: Gather Context & Friction

See `references/context-gathering.md` for detailed commands.

**Quick version:**
`