manifest.json
One machine-readable index for agents. No guessing. No scraping filenames. No “where did the OCR go?” energy.
local-first • agent-first • gloriously inspectable
Skriver turns meetings into evidence your agent can actually use.
Audio in. Video in. Notes in. Out comes a tidy run directory with a readable transcript, OCR context, diarization artifacts, confidence flags, and a manifest that tells your agent where to look next.
- Static HTML that agents and crawlers can parse without drama.
- No built-in fake reasoning layer. The artifacts do the talking.
- Readable for humans, structured for agents.
>
Quick Start
One-liner first. Confusion later. Preferably never.
Fastest path on macOS and Linux.
curl -fsSL https://skriver.ansund.com/install.sh | bash Direct global install from GitHub.
npm install -g github:ansund/skriver If you already live in pnpm land.
pnpm add -g github:ansund/skriver PowerShell install script.
irm https://skriver.ansund.com/install.ps1 | iex
Honest version: the one-liner gets the CLI onto your machine. Then skriver doctor
tells you which local media tools are still missing for full transcription.
*
For Agents
Built for the second step, not just the first command.
Skriver doesn’t pretend the transcript is the end of the workflow. It produces the follow-up bundle your agent needs: a transcript that stands on its own, a manifest that points to everything else, and a workflow guide that explains the review path.
workflow.md
A short checklist for humans and agents so the post-processing workflow stays explicit and repeatable.
llms.txt
The site ships with agent-readable product notes too, because crawlers deserve a clean path and not a JavaScript labyrinth.
//
Artifacts
The transcript is readable. The folder is inspectable. The rest is workflow.
transcript.mdReadable transcript with timestamps, notes, corrections, and speaker labels when diarization succeeds.
manifest.jsonMachine-readable index that tells agents where every useful artifact lives.
workflow.mdShort post-run checklist for agents and humans.
summary_draft.jsonDeterministic summary candidates, actions, and open questions.
low_confidence_segments.jsonSegments worth reviewing before publishing a final summary.
speaker_diarization.jsonAnonymous diarization segments when local speaker detection completes.
context_artifacts.jsonExtracted notes from slides, PDFs, docs, emails, and image inputs.
screen_ocr.tsvOCR text from video screenshots so agents can repair context-heavy transcripts.
SEO Friendly
Plain HTML, semantic sections, zero mystery meat.
Astro keeps the page fast and crawlable. Search bots, AI agents, and actual people all get the same readable content without waiting for a client app to boot.
Quirky, not precious
A terminal toybox with manners.
The design leans playful, but the copy stays direct: what it installs, what it outputs, and how an agent should use it next.
?
FAQ
Small answers for the questions people and crawlers both tend to ask.
Does Skriver do the summary itself?
Not in the hand-wavey LLM product sense. Skriver does deterministic extraction and drafting so your agent can reason from evidence instead of guesswork.
Why Astro for the website?
Because the page stays static, semantic, and crawlable. Agents and search crawlers get the actual content directly from HTML instead of chasing client-side rendering.
What happens after install?
Run `skriver doctor`, then `skriver transcribe ...`. Every run produces a clean bundle with `manifest.json`, `workflow.md`, and the transcript artifacts your agent needs.