From ai-memory
Save a session summary to memory. The "write" side of session continuity — /load is the "read" side.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ai-memory:saveThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The full transcript is saved automatically — compact captures the **substance**: what was decided, what went wrong, what the user wants. Not a play-by-play.
The full transcript is saved automatically — compact captures the substance: what was decided, what went wrong, what the user wants. Not a play-by-play.
Compact is a handoff to a future agent — write what you'd need to continue this work with no other context. Be concrete and actionable, not polished.
Two parts: a structured header for quick orientation, then a compression-gradient narrative where detail increases toward the end (recent work matters most for resumption, early work only needs enough context to understand the arc).
Skip compact for short sessions — if the compact would be comparable in size to the sum of all messages, don't generate it. Just call memory_session without the compact field; the transcript is saved automatically and is sufficient for recovery.
Goal: <why this session exists — intent, not just topic>
Status: in-progress | completed | blocked
Next: <immediate next action, one line>
Blockers: <if any, otherwise "none">
---
<Overview sentence — what was this session about, what's the outcome.>
<Early session — high compression: 1-2 sentences per topic.>
<Middle session — moderate detail: key decisions with rationale.>
<Recent work — full detail: what was tried, what worked, what didn't.
Include code snippets, exact error messages, specific function names.>
<Current state: what is working, what is broken, what is half-done.>
## Dead ends
- <approach tried → why it failed> (skip section if none)
## User requirements
- <requirements, preferences, corrections from the user>
- <specific to this project/task — things the next agent must respect>
/load budget is shared with facts and transcript tail; a bloated compact crowds them out.memory_session mcp toolPass session_id: ${CLAUDE_SESSION_ID}
Pass project: from the SessionStart context
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub dankinsoid/ai-memory --plugin ai-memory