Create Retrospective Skill for organizational learning

Home Epics Create Retrospective Skill for organizational learningEdit

Cancel

Title *

Project *

Description

## Problem

Current system has valuable work scattered across multiple sources:
- Tickets (requirements, implementation summary in working_memory)
- Memories (implementation details, patterns, errors)
- GitHub PRs (code changes, review comments)
- Proposals (ideas, decisions)

When agents need to understand past work, they must manually hunt across all these sources. More importantly, **lessons learned are not explicitly captured** - only facts.

## What You're Building

A retrospective skill that generates comprehensive learning documents by combining information from multiple sources.

### Skill Input/Output

**Input:** `ticket_id` (integer)

**Output:** Structured markdown retrospective with sections:
- Overview (1-2 sentence summary)
- What We Built (from working_memory + implementation memories)
- What Went Wrong (from comments + error memories + PR review comments)
- How We Fixed It (from follow-up tickets + fix memories)
- Links (Ticket, PRs, related tickets, related memories)
- Lessons Learned (patterns, checklists, decisions extracted)

### Data Sources to Query

1. **Ticket**: `get_ticket(ticket_id)` - title, description, working_memory, comments, pull_request_url, status
2. **Memories**: `search_memory(query: "ticket #X", limit: 20)` - all memories related to this ticket
3. **GitHub PR**: Use `pull_request_url` from ticket to fetch PR details and review comments via GitHub API (GH_TOKEN available)
4. **Proposals**: `list_proposals()` - filter for proposals with metadata.ticket_id matching
5. **Follow-up tickets**: Cross-reference from memories and comments

### Storage Options

The retrospective should be stored for future reference:
- **Option A:** Store as new memory with `memory_type: "summary"` and `ticket_id`
- **Option B:** Add to `ticket.working_memory` (may get large over time)
- **Option C:** Both - store as memory, add reference to working_memory

### Skill File Location

Create: `.claude/skills/retrospective/SKILL.md`

Follow the existing skill pattern:
- YAML frontmatter with `name`, `description`, `allowed-tools`
- Clear guidance on when/how to use the skill
- Examples of generated retrospectives

## Technical Notes

- GitHub API is accessible via `gh` CLI or curl with `$GH_TOKEN`
- Existing skills follow pattern: `.claude/skills/<name>/SKILL.md`
- Skills are auto-discovered by Claude Code runtime
- Consider the **example retrospective** from proposal evidence (ticket #128 archival system + ticket #146 follow-up)

## Acceptance Criteria

- [ ] Skill file exists at `.claude/skills/retrospective/SKILL.md`
- [ ] Skill is invocable with a ticket_id parameter
- [ ] Queries ticket details via get_ticket
- [ ] Searches memories by ticket_id via search_memory
- [ ] Fetches PR details/comments from GitHub if pull_request_url present
- [ ] Searches for related proposals
- [ ] Generates structured markdown with all required sections
- [ ] Stores result as memory (or adds to working_memory)
- [ ] Example retrospective generated for ticket #128 (the archival system case)

## Related Work

- **Evidence from proposal:** Tickets #128, #146; Memories #50, #52, #55
- **Reference skills:** `.claude/skills/memory/SKILL.md` for memory patterns