From origami
You have a **context folding** system active. It preserves your full
How this skill is triggered — by the user, by Claude, or both
Slash command
/origami:context-foldingThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You have a **context folding** system active. It preserves your full
You have a context folding system active. It preserves your full conversation history at variable resolution — every section has an always-visible self-compressed summary, and full detail is stored on disk, expandable on demand.
After compaction you'll see a fold index:
[CONTEXT FOLDING — 5 sections, 14200 tokens stored]
[F001 | FOLDED | 3200 tok | rel:0.15]
auth.mid>refac: jwt.decode>jwt.verify | tok.refresh.chain.fix | D:auth.mid.ts,auth.svc.ts
[F002 | UNFOLDED | 2800 tok | rel:0.85]
<full detail visible>
Each entry shows: fold ID, status, detail token count, relevance score, and the self-compressed summary.
| Tool | When to use |
|---|---|
unfold_section(fold_id) | You need specific code, errors, or decisions from a folded section |
fold_section(fold_id) | You're done referencing a section — free up context space |
list_folds() | See the full fold index with status and summaries |
write_summary(fold_id, summary) | Create or update a fold's self-compressed summary |
When you call write_summary, use this dense format — you are the
only reader, so maximise information density:
>refac, >impl, >fix, >add, >mod, >delauth.middleware.ts → auth.mid.ts"builds on F003"topic>action: key.details | outcome | D:filesExample:
auth.mid>refac: jwt.decode>jwt.verify | tok.refresh.chain.fix | D:auth.mid.ts,auth.svc.ts
Each fold shows its token count. Research shows LLM performance degrades well before context is exhausted, so aggressive folding is preferred:
The system enforces a tight token budget (20% of context window, max 3 simultaneous unfolds). Keep context lean - smaller context means better reasoning on what's actually there.
npx claudepluginhub dullfig/claude-plugins --plugin origamiTeaches the four operations of context engineering — Write, Select, Compress, Isolate — for managing token budgets, compaction strategies, and context partitioning to keep AI sessions sharp and efficient.
Guides managing Claude Code context window with /compact, /clear commands, auto-compaction config, sub-agents, targeted reads, background tasks, and conversation flows for long sessions.
Enforces concise responses, parallel tool execution, no redundant work, exploration tracking, and proactive context compression in every Claude Code session. Auto-applies at start.