Brain TODO

The brain stores Krisp action items as their own content_type='krisp_action_items' documents (separate from the meeting transcript). brain todo parses - [ ] / - [x] checklist lines out of those documents and prints one row per item, so the user can see open work without opening N transcripts.

For searching meeting content, see consult-brain. For ingesting other content, see ingest-brain.

Showing open items — brain todo

brain todo                                  # all open items, all sources
brain todo --source krisp                   # Krisp-derived only
brain todo --since 30                       # items from action-item docs ingested in last 30d
brain todo --closed                         # include checked-off items
brain todo --limit 50 --json                # JSON output (for piping / summaries)

Flags:

The parser is brain.todo.parse_action_items — it walks the body for - [ ] / - [x] lines, dedented, leading-bullet-tolerant. Items embedded in nested lists are picked up.

Pulling new action items from Krisp

Krisp action items aren't fetched by the brain CLI directly — the CLI prints a deferred-fetch handoff that the agent fulfills via the Krisp MCP. This pattern keeps the CLI free of live MCP knowledge.

Step 1 — Get the handoff

brain enrich --krisp-action-items                       # all meetings the CLI knows about
brain enrich --krisp-action-items --since 14            # last 14 days
brain enrich --krisp-action-items --source-id MEET123   # one specific meeting

The CLI prints:

• The Krisp MCP tool names it expects the agent to use: mcp__claude_ai_Krisp__search_meetings, mcp__claude_ai_Krisp__list_activities, mcp__claude_ai_Krisp__get_multiple_documents.
• The window/scope in plain English with a concrete ISO 8601 date.
• The exact brain ingest-stdin invocation to pipe each fetched action-item blob into.

Do NOT make up MCP parameter syntax based on the CLI's prose. Read your live MCP schema for the actual parameters each Krisp tool accepts. The CLI deliberately names the tools and describes scope in English; the agent parameterizes them from the live schema. This rule has been violated and corrected multiple times — never speculate about kwargs.

Step 2 — Call the Krisp MCP

Use the tools the CLI named, parameterized for the scope it described. Fetch each meeting's action-item content (Krisp surfaces this as one of the document kinds returned by get_multiple_documents).

Step 3 — Pipe each result into brain ingest-stdin

This part of the recipe is literal (the brain CLI owns this flag shape):

echo "<action items markdown>" | brain ingest-stdin \
  --source krisp \
  --external-id "<meeting_id>--action-items" \
  --content-type krisp_action_items \
  --title "Action items — <meeting title> — YYYY-MM-DD" \
  --date YYYY-MM-DD \
  --metadata '{"parent_meeting_external_id": "<meeting_id>"}'

Notes:

• --external-id MUST be distinct from the transcript's external-id — conventionally suffix with --action-items.
• --metadata.parent_meeting_external_id is REQUIRED for --content-type krisp_action_items (BadParameter otherwise). Use the meeting's stable Krisp meeting id.
• The action-items tag is auto-applied via _maybe_autotag_action_items — no need to pass --tag action-items yourself.

Step 4 — Report back

brain todo --since 7        # confirm the items are now visible

Tell the user: "Pulled N meetings (M action items, K still open)."

Closing an item

Brain doesn't have a brain todo close <id> command. To check off an item, edit the source action-item document directly:

brain edit <action-item-doc-id-prefix>
# In the editor, change `- [ ]` to `- [x]` on the line you want to close.

The body change re-chunks + re-embeds; the next brain todo run reflects it.

Operational notes

• brain todo is read-only. It never modifies docs.
• Items are matched by line — a single action-item doc can yield many rows.
• The parser does NOT promote items into the vault's "Related" fence or the graph — those are derived from [[wiki-links]] and metadata, not checklist markers.
• If the user asks "what action items came out of yesterday's calls" and brain todo --since 1 returns nothing, propose Step 1 to ingest first (don't fabricate items).