specwithcode
AGENTS.md4 starsChanged 12 days agoMITView on GitHub
Independent, MIT-licensed workbench for native coding agents. SpecWeave owns portable intent, claims and handoffs; the app owns sessions and control on each machine. WebSocket request -> command -> pure orchestrator -> events -> atomic sink (events, projections, receipts, outbox) -> effect worker does I/O and feeds results back as commands. Headless fixtures and live verifiers: docs/operations/checks.md. Working copies: docs/operations/worktree-cleanup.md. Runtime identity and defaults: docs/operations/development.md.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository6 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .cursor/rules/cursor-cloud.mdcCursor rule
- adversarial-reviewSkill
- contribution-triageSkill
- test-studio-appSkill
- test-studio-mobileSkill
The file
We split this AGENTS.md into the rules it gives an agent
4 rules. About 0 are common, 0 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Never open, copy or delete a running service's database (userdata/statev2.sqlite, -wal, -shm), not even read-only.
- Very specific: line 15
Run the smallest focused test plus the web and server builds your change touches.
- Very specific: line 24
Read docs/internals/effect-services.md before adding an Effect service.
- Very specific: line 26
Keep same-origin routing; never bake localhost VITE_HTTP_URL/VITE_WS_URL into a remote bundle.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
SpecWithCode
Independent, MIT-licensed workbench for native coding agents. SpecWeave owns portable intent, claims and handoffs; the app owns sessions and control on each machine.
Rules
- Public repository: never commit secrets, personal paths, machine names, customer data or private reports; keep them under
~/.specwithcodeor another folder outside the repository. - Third-party code keeps its own license. Keep
THIRD_PARTY_NOTICES.mdand the notices in Settings complete;node scripts/studio-brand-audit.mjschecks it. - In anything a person reads, say SpecWithCode, then "the app"; never "Studio". Code identifiers stay literal (
docs/studio/naming.md). .repos/holds read-only reference trees: never edit, import, package or bulk-transform them.- Production runs on Node 22.20.0 with pnpm 11.10.0; some builds need a separately scoped Node 24. Never change shared toolchains or provider installs to fix a build.
- Tests never touch live stores, credential folders or OAuth tokens: use a new private test home and a scratch repository. A check that must drive the live app creates its project with
origin: "automated"and deletes it after. - Never open, copy or delete a running service's database (
userdata/statev2.sqlite,-wal,-shm), not even read-only; usespecweave-studio db queryordb snapshot. Server code never opens those files throughnode:fs(docs/operations/live-database.md). - Browser checks run headless and save evidence to files outside the repository; a failure never authorizes a headed retry.
- Run the smallest focused test plus the web and server builds your change touches. No repo-wide tests, lint or typecheck unless asked (
docs/operations/heavy-checks.md). - Phone availability, restored native sessions and subscription capacity need real outcome evidence; a fixture does not prove them. Record what is unverified as a proof limit.
- Releases:
bash scripts/release/request-release.sh --by "<thread>"; never choose a version or mark a release stable by hand (docs/studio/releasing.md).
Architecture
WebSocket request -> command -> pure orchestrator -> events -> atomic sink (events, projections, receipts, outbox) -> effect worker does I/O and feeds results back as commands.
- Mutations use
createEnvironmentRpcCommand; protected methods go inCLIENT_GUARDED_RPC_SCOPES; the destination machine checks grants at execution. Test both the UI gate and the server refusal. - Read
docs/internals/effect-services.mdbefore adding an Effect service. - In tests, drain
OrchestrationEffectWorkerV2or await a persisted event; never sleep. - Keep same-origin routing; never bake localhost
VITE_HTTP_URL/VITE_WS_URLinto a remote bundle. - UI text, sizes, corners and colours come from the tokens in
docs/studio/design-guidelines.md. - Require manual acceptance only when the user or the increment's acceptance criteria ask for it.
Commands
| Check | Command |
| Web typecheck | node node_modules/typescript/bin/tsc --noEmit --project apps/web/tsconfig.json |
| Server typecheck | node node_modules/typescript/bin/tsc --noEmit --project apps/server/tsconfig.json |
Web build (from apps/web) | node ../../node_modules/vite-plus/bin/vp build |
| Server build | node apps/server/scripts/cli.ts build |
| One test file | node node_modules/vite-plus/bin/vp test run <file> |
Headless fixtures and live verifiers: docs/operations/checks.md. Working copies: docs/operations/worktree-cleanup.md. Runtime identity and defaults: docs/operations/development.md.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

