capture

/second-brain:capture — raw inbox producer

Hold unprocessed material in the active project's raw inbox (~/.second-brain/projects/<slug>/raw/) until the maintainer refines it into wiki nodes. Raw items are not searched — they are a staging area, surfaced as a backlog count at session start.

Map the user's argument to a raw-capture-cli action and run it. The CLI resolves the active project, stamps provenance, copies blobs, and dedups by content hash.

CLI="${CLAUDE_PLUGIN_ROOT}/mcp/dist/tools/raw-capture-cli.bundle.js"

• <path> or <url> or inline "text" → node "$CLI" capture <arg> [--node <slug>]
• pasted/piped text → node "$CLI" paste [--node <slug>] (reads stdin)
• --list → node "$CLI" list
• --discard <id> → node "$CLI" discard <id>
• --prune-processed → node "$CLI" prune-processed — delete this project's processed + discarded audit-trail items (unprocessed and malformed are kept). Opt-in cleanup, off by default: processed raw .md files are normally kept as provenance + the drain's truncation-recovery trail, and are never searched (search is scoped to ~/knowledge/wiki/), so keeping them has no retrieval cost. To make the drain prune automatically after each batch, set SB_RAW_PRUNE_AFTER_DRAIN=1.

--node <slug> records that the item is evidence for an existing wiki page (provenance only — the link is projected later when the maintainer processes it).

Relay the CLI's output. If it reports "could not resolve the active project", tell the user to cd into their project directory first.