The planner and router read only the first 120 characters of a tool's description. personalNotes started by describing the notes, so for '앞으로 나한테는 짧게 대답해줘' the planner planned a plain reply and the answer promised to be brief from now on without saving anything. Eval (qwen3.6-35b-a3b, game PC): from-now-on cases 0/6 -> 6/6, the one-off and note-shapes-answer cases still pass (15/15). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2.5 KiB
personalNotes tool spec (v2)
Each person's standing instructions for how Jarvis answers them ("앞으로 나한테는 반말로 해",
"날씨는 섭씨로만"). Product rules: docs/v2/proactive.md (제안 4). The app keeps the notes and
sends the asker's with every reply (app/src/admin.spec.md, "Personal notes"); this tool only asks
it to add, list or delete them.
Principles
- Only in v2, only the asker's own. The tool exists for a turn only when the v2 voice worker's
brain put a notes backend in the turn state (
turn_state.set_notes). Without one (v1, evals without a backend, scheduled tasks) it is not offered and refuses. The backend acts for the person asking only. - Saved only when asked to apply from now on. The description tells the model to add a note only for something the person wants from now on, never for a one-off request. Its first sentence says to save (action='add'), because the router and the planner read only the first 120 characters; when it only described the notes, the planner planned a plain reply and the answer promised "from now on I will answer briefly" without saving anything (live 2026-11-18).
- Notes shape answers, never permissions. The reply engine puts the asker's notes
(
run_reply_engine(person_notes=...)) in the system prompt inside a<person_notes>fence, labelled as the person's own preferences for how they are answered, to follow unless they conflict with the rules above, and never permitting an action or a tool. Fence tags inside a note are removed so it cannot close the block. Only the asker's notes are sent, so one person's notes never reach another person's answer. Eval:evals/test_personal_notes.py. - Raw data. The tool returns plain text for the reply model; no LLM call inside.
Schema
| action | needs | does |
|---|---|---|
add |
text |
saves it (whitespace collapsed, 1-300 characters) |
list |
the asker's notes: - id <id>: <text> |
|
delete |
id |
deletes one of the asker's notes |
action is read as its first known word. A refused request never reaches the app.
Backend
turn_state.get_notes() returns an object with act(action, **fields) -> dict (add: text;
delete: note, the note's id): {"ok": true, "item": {"id", "text"}} for add and delete,
{"ok": true, "items": [...]} for list, {"ok": false, "error": "..."} on failure. The voice worker
implements it (voice_worker/app_bridge.py): each action goes to the app as note_action with the
asker (voice_worker.spec.md, "Personal notes").