I/O

Split long-form text into atomic inbox notes. Chunks land in notes/ for later triage via /note process.

I/O

• Input: Space-separated text snippets and/or file paths.
• Output: Atomic notes at <vault>/notes/YYYY-MM-DD-<slug>.md, index patch.

Process

1. Resolve: Vault path per Skill("capture-pipeline") §1. Abort if unconfigured.
2. Parse: Split input into text vs file arguments. Read files. If images present, invoke Skill("image-capture").
3. Split: Chunk input into atomic thoughts. See references/split-rubric.md.
4. Capture: Delegate to Skill("capture-pipeline") with CHUNKS, VAULT_ROOT, SOURCE_PROJECT, TODAY, ORDER_MATTERS. Inline execution per references/inline-capture.md; agent fan-out per references/agent-fanout.md.

Agent vs Inline Decision

Chunks	Order matters?	Mode
1	n/a	Inline — run CAPTURE on main thread.
2–4	yes	Inline — chunks run in order so K can MATCH-append to K-1.
2–4	no	Agent fan-out — one agents/capture.md per chunk in parallel.
5+	no	Agent fan-out — always parallel.
5+	yes	Inline — sequential order preserved; run in order.

Order matters for numbered lists, narratives, build-on-each-other arguments.

Confirmation

Success: Captured N notes: <paths> (singular "note" when N=1). Partial failure: Captured N notes: <paths>\n\nFailed: K chunks: <reasons>. No NEW/MATCH labels, diff, or reasoning in output.