Skip to content
232 linesCodeBlameRaw

Pick any line to see why it is the way it is: the commit, the pull request and issue it came from, and what the agent was thinking.

Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays1# Contributing to g1t
2
3g1t is built the way it asks others to build: issues and pull requests on
4[g1t.sh/flagon-io/g1t](https://g1t.sh/flagon-io/g1t), required checks that prove
5a change works, and a merge queue that keeps `main` passing.
6
7## A change ships with its docs
8
9Documentation is part of the product, held to the same bar as the code.
10A pull request that changes what someone can do, see or call changes the
11docs in the same pull request:
12
13| If you change | Update |
14| --- | --- |
15| Something a person does on g1t.sh | The guide for it in `apps/docs/src/content/docs/guides/` |
16| An API route, a field, or an MCP tool | `apps/api/src/operations.rs` descriptions (they feed the OpenAPI document and the API reference), the example and notes for it in `apps/api/src/reference.json`, `reference/mcp.md`, and any guide that shows the call. Then refresh the docs' copy of the OpenAPI document with `G1T_WRITE_OPENAPI=1 cargo test -p g1t-api openapi`; `cargo test` fails until you do. |
g1t is one name: its agent's work, commits and comments show as @g1t, and nobody can claim g1t or g1t-agent17| How agents behave | `guides/working-with-g1t.md`, and `apps/web/public/llms.txt` |
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays18| Settings, limits or prices | The page that names them, and the table it is in |
19| A new feature | A section in the guide that owns it, linked from the docs home if it is a new task |
20
21The style, in short: plain sentences, second person, no marketing words,
22sentence-case headings, a numbered list for steps, a table for options, and
23every API mention with its exact route or tool name. Examples are copyable
24and real. Nothing is documented that the code does not do.
25
26Check the docs build before you push:
27
28```sh
29cd apps/docs && npm run build
30```
31
32## Icons
33
34- **Interface icons come from [Lucide](https://lucide.dev/icons).** In the
35 apps, import them from `lucide-react`. In the docs, use `@lucide/astro`:
36 import an icon component, or give `apps/docs/src/components/Card.astro`
37 and `Aside.astro` a Lucide name such as `icon="git-branch"`. Starlight's
38 own `<Card>`, `<LinkCard>` and `<Aside>` draw Starlight's icon set, so the
39 docs import the wrappers in `apps/docs/src/components/` instead.
40- **Brand marks come from [Simple Icons](https://simpleicons.org)**
41 (`simple-icons`), and follow each brand's own usage guidelines. Add the
42 dependency with the first one you show.
43- **The g1t logo and illustrations are our own artwork**, in
44 `apps/web/app/components/logo.tsx` and `art.tsx`.
45- Never hand-copy an icon's SVG paths into a file. If Lucide lacks the
46 icon you need, pick the closest one that it has.
47
CONTRIBUTING says how the site's interface is built: from the shadcn components in apps/web/app/components/ui, themed with g1t's tokens, adding a missing one there first, changing looks through variants rather than one-off class strings, and looking at every changed page in both themes on a desktop and a phone.48## Interface components
49
50- **Build the site from the components in `apps/web/app/components/ui`**,
51 which are [shadcn/ui](https://ui.shadcn.com) components themed with
52 g1t's tokens: buttons, cards, badges, tables, tabs, separators,
53 breadcrumbs, menus, dialogs, sheets, tooltips, form controls.
54- **If shadcn has a component you need and `ui` doesn't, add it there**
55 first, themed with the tokens, then use it. Don't rebuild it inline.
56- **Change how something looks through the component's variants**, not
57 with a one-off class string at the call site. A long `className` on a
58 `div` that draws a card, a pill or a button is a component that should
59 exist.
60- Hover hints use the `Tooltip` component, never `title=`. Native
61 `<select>`, file inputs, checkboxes and radios use their `ui`
62 components.
Agents have faces, and are never mistaken for people. Every agent wears a little bot face drawn from a look it owns, shape, colour, eyes, mouth, antenna, accessory and pattern, chosen in its builder and on its Profile tab with a live preview, Shuffle and a way back to the face its seed gives it; the face blinks on its own time, breathes, narrows its eyes while the agent works, shuts them asleep and bounces when it finishes, all of it still for anyone who asked for less motion. Wherever an agent shows, in chat, in a list, on a mention, on a review or a commit, its avatar carries an agent marker, and the people reading it are told so. In Chat, direct messages are two lists: People, and Agents, which also holds the agents you haven't talked to yet; a conversation with both a person and an agent in it is marked in the list, named in the conversation's header, spelled out by the composer and explained once the first time it opens. Agents keep their look in the agents service, which every service passes along. The chat and agents guides say so, and CONTRIBUTING makes the shared avatar the only way to draw an agent.63- **Every avatar is `Avatar`** (`ui/avatar.tsx`), and every agent's is
64 `Avatar` with `agent` set, or `AgentAvatar` (`components/agent-avatar.tsx`),
65 which wraps it. That draws the agent's bot face (`AgentFace`,
66 `components/agent-face.tsx`) from its `look` or its `avatar_seed`, puts
67 the agent marker in its corner and names it "…, agent" to a screen
68 reader. Never hand-draw an agent's face, a sparkle, a letter on a tile or
69 the marker at a call site: a place that shows an agent differently is a
70 bug.
CONTRIBUTING says how the site's interface is built: from the shadcn components in apps/web/app/components/ui, themed with g1t's tokens, adding a missing one there first, changing looks through variants rather than one-off class strings, and looking at every changed page in both themes on a desktop and a phone.71- Look at every page you change, in light and dark, on a desktop and a
72 phone, before calling it done.
73
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays74## Before you push
75
76- `cargo test` in the crate or service you changed.
77- `npx tsc -b --force` in `apps/web` (the incremental build misses changes
78 in `packages/contracts`).
79- Look at what you changed in a browser. Screenshots catch what type
80 checks do not.
81
Merge remote-tracking branch 'origin/claude/cool-newton-7nm8r9'82## Checking g1t against GitHub
83
84g1t runs GitHub Actions workflows as GitHub does. The Conformance
85workflow is how that is checked: `.g1t/workflows/conformance.yml` and
86`.github/workflows/conformance.yml` are the same file (a test in
87`crates/actions` fails when they differ), and each run's
88`conformance-report` artifact records what its jobs saw: which `GITHUB_*`
89and `RUNNER_*` variables were set and of what kind, every key of the
90contexts and the event and its type, how the workspace is laid out, and
91the values of outputs, files, caches, artifacts and expressions that
92should be the same everywhere. It reads no secrets and runs only by hand.
93
941. Run it on both, at the same commit: **Actions → Conformance → Run
95 workflow** on GitHub and on g1t.sh.
962. Download each run's `conformance-report` artifact, and name the two
97 files `github.json` and `g1t.json`.
983. Compare them:
99
100 ```sh
101 node scripts/conformance-diff.mjs github.json g1t.json
102 ```
103
104It lists what GitHub has that g1t lacks or has otherwise, then what only
105g1t has, and exits 1 while there are differences. A difference is a bug in
106g1t, or belongs in [What g1t can't do yet](https://docs.g1t.sh/about/limitations/).
107Add a step and a `CHECK_` value to the workflow for anything new g1t
108should match.
109
Spend caps: a monthly budget for comped workspaces and a daily breaker on what g1t pays110## Deploying
111
112Pushes to `main` deploy themselves: `.g1t/workflows/deploy.yml` runs
113`scripts/deploy.mjs`, which deploys only the parts that changed, migrations
114first. Every deployable part is listed in `deploy/stack.jsonc`; a new service
The docs folder is gone, and what it held lives where people read it: how a self-hosted g1t runs and how to deploy g1t to Cloudflare are pages on docs.g1t.sh under Run g1t yourself, and speed, rate limits and operating g1t.sh are sections of CONTRIBUTING.md; code that cited a file in docs/ now points to the page or section that covers it, or says what it means itself, and applied migrations and the runner images are left as they were.115or app goes there, and in the table on
116[How a self-hosted g1t runs](https://docs.g1t.sh/guides/self-hosting-architecture/#each-part)
117(`npm run test:deploy` says what is missing).
118[Deploy g1t to Cloudflare](https://docs.g1t.sh/guides/deploy-to-cloudflare/)
119covers the tool, the workflow, rollbacks, adding a unit and first-time
120setup.
121
122- **Migrations run before the code**, so the old code reads the new schema
123 for a minute or more. Add tables and columns; change what rows mean in
124 two deploys (code that reads both forms first); never drop what live code
125 still reads.
126- **The self-hosted runner** is released, not deployed: bump `version` in
127 `crates/runner/Cargo.toml`, merge, and push a `runner-v<version>` tag.
128 `.g1t/workflows/runner-release.yml` builds, signs and publishes it, and
129 runners update themselves to it. `scripts/runner-release.mjs` does the
130 same by hand.
preparing for cloud work131- **The desktop app** is released the same way: bump `version` in
132 `apps/desktop/src-tauri/Cargo.toml`, merge, and push a
133 `desktop-v<version>` tag. `.g1t/workflows/desktop-release.yml` builds
A desktop release goes out with what can be built. Linux is built on g1t's machines in every release; Windows and macOS each wait on a repository variable: a machine registered for them, or for Windows a cross build on Linux once aka.ms and download.visualstudio.microsoft.com are workflow-only domains, which is where the first release stopped. A platform with no machine is skipped and the download page says it is coming, with a word for the visitor whose platform this release lacks. The guide and CONTRIBUTING say so.134 the installers (Linux on g1t's machines; Windows and macOS on machines
135 registered for them, or Windows cross-built on Linux once its download
136 hosts are allowed, each switched on by a repository variable the
137 workflow's header names; a platform with no machine is left out of that
138 release), signs the updater's files and publishes them, and installed
139 apps update themselves. `scripts/desktop-release.mjs` does the
preparing for cloud work140 same by hand. The app loads the live site, so a change to the site needs
141 no release; only a change under `apps/desktop` does. The guide is
142 [The desktop app](https://docs.g1t.sh/guides/desktop/).
The phone app with your own g1t, on g1t.sh/download until Google Play, and Node 24 (#5)143- **The phone app** for Android is released on g1t.sh/download while it
144 is not in Google Play (that waits on Flagon's D-U-N-S number): bump
145 `version` in `apps/mobile/app.config.ts` when its native side changes,
146 merge, and push a `mobile-v<version>` tag.
147 `.g1t/workflows/mobile-release.yml` has EAS build the APK and publishes
148 it; `scripts/mobile-release.mjs` does the same by hand. A change to the
149 app's code alone needs no release: `eas update --channel preview` in
150 `apps/mobile` reaches installed copies. Its server half (chat scopes,
151 `/-/notify`, notify's phone push) deploys with the site and services.
152 The guide is [The phone app](https://docs.g1t.sh/guides/mobile/).
The docs folder is gone, and what it held lives where people read it: how a self-hosted g1t runs and how to deploy g1t to Cloudflare are pages on docs.g1t.sh under Run g1t yourself, and speed, rate limits and operating g1t.sh are sections of CONTRIBUTING.md; code that cited a file in docs/ now points to the page or section that covers it, or says what it means itself, and applied migrations and the runner images are left as they were.153
154## Speed
155
156Pages are a few rounds of service calls, and each round costs a trip to
157the databases, so start calls together and add no rounds.
158
159- Every page answers with `Server-Timing`: each loader, each service's
160 calls and their database time. Look at it in DevTools when a page feels
161 slow.
162- `scripts/perf/measure.ps1` times pages from your machine (`-BrowserUA`
163 for signed-out pages as a browser sees them). A page that only reads
164 must not set the `g1t_d1` cookie: a new read-only service method goes in
165 `READS` in `apps/web/app/lib/perf.ts`.
166- Targets from the US: signed-out pages under 200 ms to first byte,
167 signed-in pages under 400 ms, streamed panels within a second.
168- Workers run without Smart Placement; `scripts/perf/placement-probe.mjs`
169 measures a placement before you pin one.
A doc opens at once: the page carries the document's saved state, so the editor mounts without waiting for its live room, which connects in the background and sends only what changed, with a quiet dot that turns green when it's synced; edits made before then are kept and sent once. The editor's code is preloaded from the page and fetched early when you hover a doc, the app's shared code ships in a few chunks instead of a hundred small ones, socket tickets come with the page instead of a round trip, and the artifacts service answers a page, a sidebar and a room in parallel lookups instead of a dozen in a row, reporting its database round trips in Server-Timing. The artifacts guide says how a doc opens, and CONTRIBUTING's Speed section has the numbers.170- A TypeScript service reports its database time too (`db;dur`, from
171 `openD1` in `@g1t/contracts`): the round trips a method makes are what
172 it costs, so start them together (`Promise.all`), keep what never
173 changes across requests in the isolate (the artifacts service keeps a
174 workspace's slug for 30 s and whether its General space exists), and
175 don't wake a Durable Object for a read D1 can answer.
176- Nothing live gates a page. A page carries what it shows (a doc's text
177 and its saved document), opens its sockets as it hydrates rather than
178 from inside the code they feed, keeps what is typed until they answer,
179 and says so with a dot, never a spinner. A page opened with an access
180 token gets its socket tickets with the page (the root loader mints the
181 feed's, an artifact's loader its room's) rather than asking for each.
182- Code that only the browser runs (the doc editor) is still fetched with
183 the page: `vite.config.ts` notes the chunk and what it imports for the
184 route's `links`, and a link to such a page warms the chunk on hover.
185 Modules several routes share are grouped by who shares them
186 (`codeSplitting.groups`), so a page fetches tens of files, not a
187 hundred and more. Before this, opening a doc fetched 124 files
188 (1,049 KB) in three waves and the editor appeared at 2.8–3.3 s; a doc
189 now fetches 71 files (1,026 KB) in one wave, and against the deployed
190 services the editor appears at 2.1 s, of which the room's answer is
A doc saved before g1t kept documents with the page opens as fast as any other from its second open: the page asks its room for the document once, sends it with the page and keeps it on the row, instead of leaving the editor to wait for the room's first message. The artifacts guide says so, and CONTRIBUTING's Speed section has today's measurements on g1t.sh and what is still to cut.191 0.65 s and the editor's own start 0.4 s. With the saved document in the
192 page, measured on g1t.sh signed in (2026-10-10, warm, 1440×900), the
193 editor appears at 1.5 s: first byte 0.53 s, scripts by 0.75 s, the
194 editor's own start the rest. A doc without a saved document (one last
195 saved before documents were kept) waited for its room's first message
196 instead, 0.9 s after the socket opened, and appeared at 2.0–2.5 s; the
197 page now asks the room once and keeps the answer, so only its first
198 open pays. Still to cut: the live socket's four service calls before
199 the room answers, the page and sidebar's 27 database round trips in two
200 calls, the root loader's billing and identity calls (0.7–1.3 s and
201 0.45 s inside), and the 3.4 MB of script a doc page decodes.
The docs folder is gone, and what it held lives where people read it: how a self-hosted g1t runs and how to deploy g1t to Cloudflare are pages on docs.g1t.sh under Run g1t yourself, and speed, rate limits and operating g1t.sh are sections of CONTRIBUTING.md; code that cited a file in docs/ now points to the page or section that covers it, or says what it means itself, and applied migrations and the runner images are left as they were.202
203## Rate limits
204
205`RATE_LIMITS` in `packages/contracts/src/rate-limits.ts` is the table of
206record, and the [rate limits page](https://docs.g1t.sh/reference/rate-limits/)
207is the public one: change both, and the binding in the Worker's
208`wrangler.jsonc`, together (`front-door-limits.test.ts` fails when they
209disagree). Limits fail open, keys hash anything secret, and each Worker
210takes its namespace ids from its own block of a hundred (41xx packages,
21142xx web, 43xx repos, 44xx api, 45xx og, 46xx status).
212
213## Operating g1t.sh
214
215For Flagon staff.
216
217- **Incidents** are declared, updated and resolved in sudo, under
218 **Platform → Incidents**, and appear on status.g1t.sh within 30 seconds.
219 Declare as soon as people are affected, and pick the higher severity
220 when unsure. SEV1 (down for most people, or data at risk) gets an update
221 at least every 30 minutes and SEV2 (a core part broken for many) at least
222 hourly; both email subscribers and need a blameless postmortem within
223 five working days.
The artifacts service is services/artifacts, the Worker g1t-artifacts, bound as ARTIFACTS by the API, the site and the agents; its live rooms move to it with a Durable Object transfer from g1t-docs-service, and its database, bucket, indexes and queue keep their names. The git store's binding and settings are GITSTORE, its ops scripts gitstore-*, and workflow run artifacts keep their compatible API under run_artifacts modules. The deploy tool puts a Worker that has never deployed before the Workers in its stage that bind to it, and the deploy guide gives the cutover runbook.224- **An Artifacts outage:** `scripts/ops/gitstore-namespaces.mjs` shows a
The docs folder is gone, and what it held lives where people read it: how a self-hosted g1t runs and how to deploy g1t to Cloudflare are pages on docs.g1t.sh under Run g1t yourself, and speed, rate limits and operating g1t.sh are sections of CONTRIBUTING.md; code that cited a file in docs/ now points to the page or section that covers it, or says what it means itself, and applied migrations and the runner images are left as they were.225 failing namespace. After 15 minutes, serve it read-only from the fallback
226 git store (`scripts/ops/restore-to-gitstore.mjs restore`, then the repos
227 Worker's secret `GIT_FALLBACK_NAMESPACES`) and open an incident. To
228 switch back, `reconcile` anything the fallback took, then delete the
229 secret. The script's header has every step.
230- **Costs and margin**, model prices, credits and the platform pause are
231 in sudo, under **Costs & margin**. The code is in `services/billing/src`
232 (`costs.rs`, `margin.rs`, `pricing.rs`, `budget.rs`, `platform.rs`).

This file's history is long; its oldest lines are credited to the oldest commit read.