The Daily Token
A daily digest of what landed in your repos: agents write it in plain words, cut it into short animated reels, and every claim is checked against the data.
What you get
Agents open more PRs than anyone can follow. Each issue turns a day of merges into four tabs:
- Digest: one thread of 6–12 s animated shorts per story, played like stories on a phone. Tap for the next one.
- Ledger: one card per PR, in plain words, with the technical detail under the hood.
- Blueprint: a map of the system, with the parts that changed today lit up.
- Cutaway (optional): a narrated 40–90 s tour of how one part of the system works.
On the weekly day, the issue also looks back at the week.
How it works
collect ─▶ stories ─▶ reels ─▶ motion ─▶ check ─▶ render ─▶ publish
GitHub Writer Director Explainer
⇄ Editor ⇄ Readers ⇄ Viewer
- Code collects the day from GitHub.
- Agents write the stories and reels. Other agents review them against rubrics, and the author revises until the review passes.
check refuses any PR, count or quote that isn't in the collected data.render builds the pages. publish is optional.
Your digest is its own repo, the data root: project.json, a map of the system, and one folder of JSON per issue. Every publish renders every issue again from its JSON, so a fix to the look reaches the whole archive.
Get started
With Bun installed, run any command as bunx daily-token …, or install it once with bun add -g daily-token. A digest made by init pins the engine in its own package.json. To work on the engine itself, clone this repo, bun install, and use bun src/cli.ts wherever this README says daily-token.
daily-token preview # the made-up sample issue on :5173
daily-token init my-digest # asks about the project, then offers to make its GitHub repo and set its keys
cd my-digest && export GH_TOKEN=... # read-only token for those repos
daily-token project cartographer # proposes a map of the system
daily-token project cartographer --accept # keeps it
daily-token project brief --accept # optional: names and terms from the repos' docs
daily-token run --date 2026-10-20 # make an issue
daily-token preview # read it on :5173
project brief writes a proposal first; run it without --accept to read or edit it before you keep it.
init asks for the repos, time zone, cutoff, issue days, stories, the Cutaway, who may read and the name on the share cover. In a terminal it then offers, step by step and each skippable, to make the digest's private repo on GitHub, set its Actions secrets and variables with gh (it says where each key comes from), and put the site online. init --dry-run prints those commands, values hidden, and runs none.
Hosting is optional. To share issues, daily-token github-app makes the GitHub App readers sign in with, and daily-token deploy puts the bucket and the site on Cloudflare. Readers sign in with GitHub, and whoever can see every one of the project's repos may read it. init also writes a GitHub Action that makes the issue every issue day. Any S3-compatible bucket with daily-token serve works too. See docs/deploy.md.
Requirements and cost
- Bun 1.3+.
- The Claude Code CLI, logged in.
- Optional: a Cloudflare account to host the site.
The agents run on your Claude login. Real issues have cost about $2–12 of model time each.
Configuration
| Setting | Where | Default |
|---|
ANTHROPIC_API_KEY | env (CI only) | your Claude login locally |
GH_TOKEN | env | none: a read-only token for the repos |
DIGEST_TZ, CUTOFF | env | UTC, 00:00: when the day ends |
days | project.json | every day: the days an issue comes out |
weekly | project.json | none: the day an issue also recaps the week, e.g. "fri" |
pipeline | project.json | rounds, budget and reviewers per stage, and the most stories an issue may have (stories.maxStories, 6); see docs/pipeline.md |
Documentation
- Commands: every command and its flags
- The pipeline: what is collected, the agents and the review loop
- An issue: the files in a data root, the tabs, and formats
- Templates: reel templates, scenes and animation marks
- The Cutaway: the narrated tour and what it needs (ffmpeg, Node, Chrome)
- Deploy: workflows, the bucket,
serve, the Cloudflare Worker, sign-in
License
Apache-2.0. The starter files init copies into your data repo are MIT.
