Recover vanished meeting commitments with owners and drafts you send
Mine recent meeting minutes for **said-once, never-seen-again** commitments, lock each row to a **named owner** and a **next human step**, and generate **calm nudge drafts you paste yourself** so follow-through stays fast without surprise pings.
Plus: three prompts — **vanish scan** across weeks of minutes with quotes only, **one sendable nudge per stalled row** (you send it), and **dedupe / hand-off** when you also keep a cross-meeting ledger.
| Audience | Everyone · Executives |
|---|---|
| Time | ~10 min |
| Prerequisites | You can attach context files and chat with an agent ([Add your first context file](/tutorials/add-your-first-context-file)). Strongest if you already extract action items from single meetings ([Turn meeting notes into action items and decisions](/tutorials/turn-meeting-notes-into-action-items)). If you keep rolling memory across weeks, pair this lesson with [Build a cross-meeting commitments ledger](/tutorials/cross-meeting-commitments-ledger) and use the overlap section below so nothing fights itself. |
| You'll end up with | A **Commitment follow-through** (or similarly named) agent chartered for quote-grounded vanish scans plus nudge drafts, your minutes attached as context, and one completed pass where at least one stalled commitment has a named owner and a draft message **you** chose to send or skip. |
When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.
Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.
Why this matters
Minutes are full of good intentions. Follow-through is where promises go to wait. A commitment that sounded decisive on Tuesday can disappear by Friday because nobody’s job is circle back on every thread we ever opened.
Framing the problem as follow-through + owner + nudge keeps it operational: you are not scoring personalities. You are asking who owns the next step, what happened since we said it, and what one calm sentence would unblock it. Auxot drafts the language and finds the drop-offs in the text you already filed. You send the nudge so trust stays intact.
Today you charter one agent for that loop. The next time a commitment goes quiet, you already have the owner lined up and a draft sitting in your clipboard instead of a guilty vague Slack.
The agent reads the minutes you attach. It does not attend meetings for you.
How this overlaps other meeting tutorials
Overlap managed so you do not build three conflicting systems:
| You already use | Add this lesson for |
|---|---|
| Turn meeting notes into action items and decisions | After single-meeting cleanup: run a vanish scan across the last few weeks so commitments that never came back get an owner and a nudge draft. |
| Build a cross-meeting commitments ledger | Use the ledger for memory (OPEN / STUCK / DONE). Use this lesson’s prompts as the weekly nudge pass on OPEN rows, or run a vanish scan first, then paste new rows into the ledger so reconciliation stays one-way (minutes → ledger). |
If you run both ledger reconciliation and this nudge pass weekly, keep one rule: same quote standard. No quote, no new commitment, no nudge.
Quick start
- Gather minutes as context files — Settings → Context Files → Add — one file per recurring meeting type is still the strong pattern (same advice as Add your first context file). At minimum, attach the last three to four sessions you care about.
- Open Chat with the Admin Agent — click Chat in the left menu — ask it to create a Commitment follow-through agent with this charter: read attached minutes, find commitments that do not reappear in later notes, never invent text, require a short quote for every flagged line, output Owner (explicit or
AMBIGUOUS), Source meeting + date, Next human step (one sentence), and Nudge draft (one short paragraph, calm tense, ends with an ask; do not send). - Attach those context files to the agent and lock the rule: no quote, no flag.
- Run the vanish-scan prompt from The agent can do that? → 1 below. Review each row: fix wrong owners before anything gets sent.
- Paste one nudge you actually stand behind (or delete the row if the commitment is dead). Optional: copy stable rows into your ledger file if you use one.
Done? The agent is saved under Settings → Agents, and you have one thread where a stalled commitment moved because a human sent one clear message.
The agent can do that?
1. Vanish scan across weeks (quotes required)
Open Commitment follow-through (or the Admin Agent while you iterate):
Attached files are meeting minutes in chronological order (filenames show dates). Task: (1) List commitments or strong "we should / I'll / we'll" lines that appear in EARLY weeks but have NO echo in LATER weeks (echo = same topic, owner update, or completion). (2) For each flagged item output: Quote | Inferred commitment (one line) | Last meeting it appeared | Owner (name or AMBIGUOUS) | Why follow-through likely stalled (neutral, max two sentences). If you cannot quote the original statement, omit the row. End with OPEN_ROWS: a markdown table: Commitment | Owner | Source | Stall read | Next human step.
Why it’s non-obvious: Models love inventing “broken promises” from tone. The echo rule forces evidence in the text, not in team drama. You get a list you could defend in standup.
2. One nudge draft per OPEN row (you send it)
OPEN_ROWS: [paste table]. For each row with Owner not AMBIGUOUS, draft exactly one Slack or email paragraph: opens with the commitment in plain language, names the owner, cites the source meeting date, proposes one concrete ask or deadline, ends with "Reply if this is already done or I'm wrong on ownership." Do not send anything; output drafts only.
Why it’s non-obvious: Nudge only works if you press send. Autonomous chasing burns trust. The draft’s job is to remove the blank-page delay, not to automate guilt.
3. Hand-off when you also keep a ledger
If you maintain Build a cross-meeting commitments ledger:
Compare OPEN_ROWS to the Commitments ledger context file. Propose: (1) MERGE if a vanish-scan row matches an existing ledger row, (2) ADD only if the quote is new, (3) STATUS flip to STUCK if the nudge draft is the third ping with no movement. Output ledger-ready markdown tables only; no new rows without quotes.
Why it’s non-obvious: Without a merge step, you duplicate rows and people get two pings for one promise. The agent is better at diffing structured tables than humans are under time pressure.
Go deeper
When ownership stays AMBIGUOUS
Do not send the nudge. Update the ledger or minutes notes after the next meeting with explicit assignment, then rerun. Half of follow-through is fixing the capture, not nagging.
Troubleshooting
- Too many flags: Narrow the window (two weeks, one team) or require two missing echoes before a flag.
- People ignore nudges: Shorten to one sentence plus deadline; drop preambles.
- Ledger and this agent disagree: Trust quoted text. If the ledger row has no quote lineage, treat ledger as hypothesis until minutes back it.
Walkthrough
Step 1: Name the agent and lock the hard rules
In chat with the Admin Agent, say you want a Commitment follow-through agent. Include:
- Attached meeting minutes (context files) are the only source of truth.
- No quote, no row.
- Outputs are drafts for humans to send, never auto-sent.
- Status vocabulary stays small: use OPEN, STUCK, DONE, or AMBIGUOUS for ownership only.
Save the agent under Settings → Agents.
Step 2: Run your first vanish scan
Attach the same files you listed in the charter. Paste the prompt from The agent can do that? → 1. Read OPEN_ROWS like a finance exception list: if you would not defend a row in front of the group, delete it.
Step 3: Generate nudges for the rows you stand behind
Take the filtered table. Paste into The agent can do that? → 2. Edit voice to match your culture (shorter, warmer, fiercer). Send one message yourself. Watch whether the thread moves.
Step 4 (optional): Sync with a ledger
If you use a ledger file, run The agent can do that? → 3 so vanish-scan rows land once.
Next steps
- → Split vanity metrics from decision metrics when follow-through rows need the same honesty rubric as exec decks — quarantine vanity before it crowds the slide.
- → Build a cross-meeting commitments ledger when one-off nudges are not enough and you need rolling OPEN row memory.
- → Turn meeting notes into action items and decisions if you are not yet getting structured decisions and owners out of each meeting.
- → Keep your context files honest and fresh so stale minutes do not invent fake drift.