reflect

SKILL.md

Reflect

Capture session learnings before ending. Counterpart to /hydrate.

Workflow

  1. Review what happened this session:

    • What did we work on?
    • What decisions were made?
    • What did we learn?
    • What's unfinished or needs follow-up?
  2. Identify memories worth preserving:

    • Episodic: Significant events, breakthroughs, frustrations
    • Semantic: New facts learned, definitions clarified
    • Procedural: Patterns discovered, how-tos established
    • Strategic: Decisions made, approaches chosen, plans formed
  3. For each memory, assess:

    • Type (which folder)
    • Importance (0.0-1.0)
    • Concepts (search keywords)
  4. Create memory files in <vault>/Areas/AI/Memory/{Type}/:

    • Filename: YYYY-MM-DD - Brief Title.md
    • Use standard frontmatter format
  5. Update living context if significant changes:

    • Major state change -> Update Areas/AI/Context/Current State.md
    • New decision -> Update Areas/AI/Context/Decision Register.md
    • New connections -> Update Areas/AI/Context/Connections Map.md
  6. Optionally create session log in Areas/AI/Collaboration/Sessions/:

    • Use existing session log template format
    • Link to memories created
  7. Summarize what was captured

Memory File Template

---
created: {today}
type: {episodic|semantic|procedural|strategic}
importance: {0.0-1.0}
concepts: [{keywords}]
source: session
---

# {Title}

{Content with context}

## Context

{Why this matters, when it applies}

Reflection Prompts

Ask yourself:

  • What would I want to know starting fresh on this topic?
  • What mistakes shouldn't be repeated?
  • What worked well?
  • What context would be lost if not stored?

What to Capture

Always capture:

  • Decisions and their rationale
  • Discovered patterns or processes
  • Project context that took time to establish
  • Unfinished work and next steps

Consider capturing:

  • Interesting technical details
  • User preferences observed
  • Problems encountered and solutions found

Skip:

  • Routine operations with no novel learning
  • Information already well-documented elsewhere
  • Transient details unlikely to matter later

Example

User: /reflect

Response: "Reflecting on this session...

We worked on: Obsidian-native memory system design

Key memories to store:

  1. Strategic: Decision to replace Hexis with Obsidian files
  2. Procedural: Memory file format with frontmatter schema
  3. Semantic: Commands live in ~/.claude/commands/

Creating 3 memories... -> Areas/AI/Memory/Strategic/2026-01-04 - Replace Hexis with Obsidian.md -> Areas/AI/Memory/Procedural/2026-01-04 - Memory File Format.md -> Areas/AI/Memory/Semantic/2026-01-04 - Command Location.md

Session context preserved. 3 memories stored."

Parameters

If $ARGUMENTS provided, use as focus area for reflection. Otherwise, reflect on the full session.

Response Format

  • Brief summary of session work
  • List of memories being stored (with types)
  • Confirmation of files created
  • Any notes about context updates or unfinished items
Weekly Installs
6
First Seen
Jan 23, 2026
Installed on
claude-code5
opencode3
gemini-cli3
windsurf3
codex3
cursor3