Hand your meeting notes to Claude
Most meeting tools end at the summary. You leave a call, the tool spits out three bullets and an action-item list, and that's the product.
Fieldnotes ends at the transcript. On purpose. A speaker-labeled transcript with timestamps is the source of truth — the offhand commitment, the disagreement that mattered, the actual moment somebody changed their mind. A summary is a useful artifact, but if you don't have the source you can't go back and check it when it matters.
The bet I made when I built Fieldnotes: most of you already pay for a real AI. You don't need a second subscription to summarize your meetings. Hand the transcripts to the agent you already have.
Here's what that actually looks like.
Use the agent you already pay for
Two ways this works today, depending on how you like to talk to Claude.
Cowork is the easier one if you don't live in a terminal. You open Claude, point it at the folder where your Fieldnotes meetings land, and it can read, summarize, search, and draft against the whole thing the same way it would help you with a project. It's all in the same chat window you already use.
Claude Code is the one I use, because I already have it open all day. It's Anthropic's command-line version of Claude — runs on your laptop, reads your files where they sit, and you can type a single command to summarize a folder full of meetings at once.
Either way, the trick that makes this nice is that you install a small recipe called a skill. The skill teaches Claude the shape of your Fieldnotes exports — markdown files, one per meeting, with speakers and timestamps — and gives you a handful of commands you can run by name. Without the skill you'd be pasting one transcript at a time and re-explaining yourself. With it, you say /process-fieldnotes and Claude knows what to do.
The skill I'm using here is a single file. Install it once. The full file lives at fieldnotes.fm/skill — copy it into the right folder for your setup and Claude picks it up the next time you open it.
There's also the privacy story worth saying out loud. Your transcripts never leave your machine when Fieldnotes makes them. When you process them this way, the meeting files stay on your disk — Claude reads them in place. Compared to most meeting tools, where the recording, the transcript, and the summary all live in somebody else's database, this is a meaningful difference.
/process-fieldnotes — summarize everything new
This is the one I run most. It walks your meetings folder, finds every meeting that doesn't already have a summary, and writes one next to it.
> /process-fieldnotes
Processed 4 new meetings. Skipped 12 already-summarized.The summaries follow a fixed shape: what we decided, action items with owners, open questions, quotes worth keeping, loose ends. The skill is opinionated about one thing — if the "decisions" section is just a paraphrase of what people said, the summary failed. Decisions are commitments with consequences. If a meeting had no real decisions, the summary says so instead of manufacturing them.
The "skipped" part is the trick that makes this cheap to run on a cadence. You can open Claude on a Friday afternoon, type /process-fieldnotes, and it only works on what's actually new.
/fieldnotes-recap — synthesize across meetings
Summaries per meeting are useful. Synthesis across meetings is where it stops feeling like a tool and starts feeling like leverage.
> /fieldnotes-recap this week
# Recap: this week
**Meetings:** 7 across May 11 – May 15
## Thread: Reskin direction
The team converged on Field Notes National Parks as the visual target —
kickoff Monday, alignment Wednesday, copy implications Thursday.
## Thread: API contract for vendor X
Open from Monday's planning, still open Friday. Jeff pushing for the
strict version, Bob arguing for a softer cutover.
## What's now decided
- Ship the reskin in two phases
- ...
## What's still open
- Vendor X cutover style
- ...The scope can be a date range, a person's name, or a topic. "This week", "last week", "Jeff", "pricing", "May 11 through May 15" — they all work. The skill reads the per-meeting summaries first if they exist, and falls back to the full transcripts when it needs the actual quotes.
The discipline I tried to encode: themes have to emerge from the text. If the meetings don't share a clear theme, the skill says so honestly instead of inventing one. There is nothing more useless than a synthesis that invents a narrative.
/fieldnotes-search — find what was actually said
The transcript-as-source-of-truth thing pays off here. When somebody emails you next Wednesday asking "wait, did Jeff actually commit to that?", you don't have to remember. You search.
> /fieldnotes-search "shipping date"
## 2026-05-12 Product Sync.md
> **Jeff** _12:34_ ... we should commit to a shipping date by end of week ...
## 2026-05-09 Roadmap Review.md
> **Ryan** _8:21_ ... the shipping date for v1 is still aspirational ...Speaker, timestamp, surrounding context. The skill is deliberately not aggressive about truncating — you're searching because you want to see what was actually said, not a paraphrase of it.
/draft-followup — turn a meeting into an email
The last one is the one that saves me the most time per use, even though I run it less often.
Give it a meeting (fuzzy match: "yesterday's design review", a title substring, whatever), and it drafts a follow-up email in your voice — recap, action items with owners, open questions, sign-off. If you want it to sound more like you, drop a short voice profile into the skill folder and it'll use it. If not, it defaults to neutral professional.
It outputs the draft as a copy-pasteable block. It doesn't send the email. I don't trust any agent enough to send email on my behalf yet, and you probably shouldn't either.
What you keep
The reason I like this loop is that the source stays yours. The transcripts live on your disk in plain markdown. The summaries live next to them as plain markdown. Nothing's in a database somewhere you don't control. If you stop using Fieldnotes tomorrow, you still have the corpus. If you switch from Cowork to Claude Code or vice versa, the skill is one file — bring it with you.
The agent does the work. You keep the source.
— Ryan