| name | swarm-checkpoint |
| description | Checkpoint the current conversation back to its Swarm channel. Use ONLY when the user explicitly asks to checkpoint, share, or save their work to Swarm — never invoke automatically, because sharing is gated on the channel's capture mode and the user's intent. |
Checkpoint to Swarm
Checkpoint the current conversation to its Swarm channel, following the guidance field returned by get_context_for_injection:
- Auto — call
share_session with trigger=agent_auto.
- Propose — ask the user first; if they agree, call
share_session with trigger=user_request.
- Off — keep it private; do not share unless the user explicitly asks.
Write it for a future reader
You hold this conversation; the server never sees the transcript, so what you write IS the record. Author the content as a faithful, complete account of what actually happened — the decisions and their rationale, what changed, what was learned — not a one-line stub. A thin capture produces thin (or empty) record suggestions, so the value you bank is the value you write. summary is the injection-ready one-liner; content is the full account.
Then check for canon
A checkpoint preserves the conversation; the durable knowledge inside it should also become records. After sharing (and even when the mode is Off), scan what this conversation produced for record-worthy canon:
- decision — did you make a choice with a rationale? (e.g. "we will lead with X because Y")
- standard — did you set a rule future work must follow? (e.g. "marketing copy must show a human moment before product language")
- fact — did you establish a durable truth, or one that extends or contradicts an existing record?
If a conversation cleared that bar, name the specific record(s) and offer to promote them with promote_context_to_record (it keeps provenance back to this conversation). Recognizing canon is proactive; creating it is not — offer good candidates and let the user decide. Skip the offer when the capture mode is Off, or when no human is present.
