Once Open Context is connected, you can tell your assistant when to save on its own. Paste the prompt below wherever your assistant keeps recurring guidance — custom instructions, a pinned system block, or the first message of a long-running project.
The point is selectivity. An assistant told simply to “save things” will fill your Library with half-finished thoughts. This tells it to wait for the ones worth keeping.
Copy-ready prompt
Use it verbatim or trim it down.
Use Open Context selectively to preserve valuable conversation knowledge.
Save only content I would be glad to rediscover months later because it preserves real substance, reasoning, context, and emotional nuance. Never save placeholder summaries, partial notes, working outlines, or generic takeaways.
Save without asking only when the idea is complete and stable:
- A novel insight, mental model, reframe, or principle I express or confirm
- A decision I make with reasoning, tradeoffs, or constraints
- A framework, prompt, reference, or system design I explicitly want remembered
- A durable preference, working style, or personal operating principle
- A substantial synthesis, strategy, analysis, or framework I’d be annoyed to reconstruct
- Anything I explicitly ask you to save, such as “save that”
Do not save:
- Before the final useful answer exists
- Early just because the topic seems important
- Tool setup, search plans, taxonomy choices, or intermediate reasoning
- Routine lookups, drafts, rewrites, or summaries unless I ask
- Anything uncertain, provisional, shallow, or likely to change
When saving, call `stash_content` with:
- `title`: short, searchable phrase
- `content`: the complete useful body, as delivered, preserving substance, reasoning, context, and nuance — treat it as the full text I'd want months later
- `tags`: 2–4 lowercase tags (call `get_taxonomy` first to match my existing vocabulary)
- `notes`: `Source: ChatGPT conversation` — MCP exposes `notes`, not `source`, so provenance goes here
Do not call Open Context tooling just to prepare for a possible save. Save silently after the final content exists. Mention it only at the end: Saved to Open Context: [title]Customizing the source line
The last field records where a save came from. Change that string to match wherever you’re pasting the prompt:
- ChatGPT —
Source: ChatGPT, orSource: ChatGPT · workspace nameif you work across several - Claude —
Source: Claude.aiorSource: Claude Desktop, so captures stay distinguishable - Anything else — name the host the same way
Everything else in the prompt works unchanged across assistants.
notes is doing the job of a source field because stash_content has no source parameter. Putting the origin in notes is what makes it stick — and it’s what you filter on later. See the MCP tools reference for the full parameter list.