Consolidating Learnings

Capture what was learned, update memory, and craft skills when patterns emerge.

Boundaries

This skill MAY: read/write .unipi/memory/, read session context, read plans/specs, write skill files if user approves. This skill MAY NOT: edit production code, run tests, deploy.

Command Format

/unipi:consolidate <string(greedy)>(optional)

• string(greedy) — optional focus (e.g., "focus on auth patterns" or "summarize what we learned about testing")
• Two modes: end-of-work (memory + registry hooks) or start/middle (read context, consolidate ideas)

---

Mode 1: End of Work Session

Triggered when run after /unipi:review-work marks work as done.

Phase 1: Gather Learnings

1. Read session context — what was discussed, decided, built
2. Read the plan and spec — what was the goal, what was achieved
3. Identify key learnings:
   • Patterns discovered
   • Decisions made and why
   • Problems encountered and solutions
   • Things that would be done differently
   • Reusable approaches

Exit: Learnings identified.

Phase 2: Update Memory

1. Check if @unipi/memory extension is installed
2. If not installed → skip memory, note to user
3. If installed:
   • Read existing .unipi/memory/ files
   • Find relevant memory files (by topic, date, or tag)
   • Update in place — don't always create new files
   • Merge new learnings with existing knowledge
   • Prevent stale data by updating, not appending

Memory file format:

---
topic: {topic}
updated: YYYY-MM-DD
tags: [tag1, tag2]
---

# {Topic}

## Key Learnings
- {Learning 1}
- {Learning 2}

## Patterns
- {Pattern description}

## Decisions
- {Decision} — {Rationale}

Exit: Memory updated.

Phase 3: Skill Crafting

1. Check if @unipi/registry extension is installed
2. If not installed → skip, note to user
3. If installed, assess if learnings are reusable:

Auto-create skill if:

• Pattern will definitely be used in future runs
• Solution applies to recurring problem
• Workflow could be standardized

Ask user if uncertain:

"I discovered a pattern that might be worth capturing as a skill: {description}. Should I create a skill for this?"

Skip if:

• One-off solution, unlikely to recur
• Too specific to current context
• User declines

Exit: Skill created or skipped.

Phase 4: Summary

Report to user:

• What was saved to memory
• What skills were created (if any)
• Suggest next steps if any work remains

---

Mode 2: Start / Middle of Session

Triggered when run without prior work session context.

Phase 1: Read Context

1. Read session conversation so far
2. OR read latest brainstorm/spec/plan
3. Understand current state — what's been discussed, what's decided

Phase 2: Consolidate Ideas

1. Summarize key points from context
2. Identify open questions
3. Identify decisions made
4. Identify next steps

Phase 3: Write Summary

Write consolidation to .unipi/memory/ — same format as Mode 1.

Phase 4: Present

Present summary to user. Ask:

1. Continue to brainstorm — if ideas need formalizing
2. Continue to plan — if decisions are clear
3. Done — summary captured, return later

---

Notes

• Memory files are living documents — update, don't always create new
• Skill creation is opportunistic — only when pattern is clearly reusable
• Both modes write to .unipi/memory/ — consistent location
• Respects extension availability — graceful degradation if extensions not installed