capture-session Card

Capture Session Cards

A capture is a user-recorded batch of photos and/or voice, delivered to chat as a <capture doc="..."> message (or, for bbx scan-import batches, dropped straight into _content/inbox/ with no chat message at all — these instructions apply wherever the card is found). The session card groups the images, audio clips, and uploaded files from one recording session; its child cards live inside the session's attach scope ({basename}.attach/), refs using the attach/ virtual prefix.

Frontmatter:

• status — new (just written, not yet annotated) → delivered (the chat message went out) → annotated (you've done your annotation pass and committed it). Older cards may carry transcribing/transcribed/intake-complete/extracted — legacy values from a retired pipeline; treat those cards as leftover new work.
• session-id — links back to the capture session.
• time — { start, end?, duration? }.
• images / audio-clips / files — manifests of the child card refs.
• source — present when the session came from outside chat: scan-upload/<token-name> means the desktop scan uploader sent these files under that named credential. It tells you which device/profile produced the scan when a batch looks wrong.
• partial — the recording cut off unexpectedly (crash, disconnect, abandonment). Treat the final seconds of transcript as possibly mid-thought — the tail may be missing, not the person trailing off.
• transcription-failed — one or more clips still need transcription (the transcription provider was unavailable when this was prepared). The capture was still delivered rather than held hostage to the outage; a later bbx transcribe/HQ pass can fill in the missing text.

Body — the assembled transcript, a timeline of transcribed speech interleaved with:

• {% image ref="attach/photo-001.image.card" /%} — where a photo was taken (description/filename come from the referenced image card).
• {% silence duration="15s" /%} — gaps of 10+ seconds.

This body is generated, not hand-written — don't edit it directly; if something needs correcting, fix the source (a child card's transcript/description) and re-derive, or note the correction in your own annotation instead.

Your duties on a capture card, in order:

1. Annotate by default. OCR any images with text and add descriptions for the rest, via subagents reading the actual image files (don't invent content from the transcript alone) — write the results onto the child image cards, commit, and set this card's status to annotated.
2. Then file it. Either bbx mv the card (and its attach scope) out of tmp-capture/ to wherever it belongs — a project, a person, a memo, a record — or, if you can fully process the capture on the spot (e.g. it's just a quick note that becomes a todo), complete that work and delete the card instead of filing it.
3. tmp-capture/ must not accumulate. It's a landing zone, not storage — a capture left there is unfinished work. If you can't finish filing it in one turn, say so and come back to it; don't leave it silently.

The contains: field

Give this card a one-sentence contains: — the prime retrieval field for bbx search and listings. How to write a good one (carry the information when it's concise, never a list of parts, under 200 characters) is in the agent guide's ABOUT_CARDS section.