ai-workflow / packages
AGENTS.md18 starsChanged 58 days agoMITView on GitHub
Status: current Last-verified: 2026-09-25 Workspace packages shared by the worker and the dashboard. contracts holds cross-application shapes and constants; conditions evaluates predicates; costs prices provider usage; harness owns model policy and built-in compatibility profiles; prompts owns prompt composition; skills owns browser-safe product skill contracts and validation; workflow-graph owns the pure rules of a workflow definition; and agent-visibility owns the record of what one send gave a model and the clarification rounds, and never composes a prompt. These pure packages may…
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 31,439 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 repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- packages/CLAUDE.mdCLAUDE.md
- artur-releaseSkill
- gate-ladderSkill
- init-agentSkill
- init-envSkill
- init-jiraSkill
- init-neonSkill
- init-slackSkill
- init-vcsSkill
- new-integrationSkill
- ai-workflow-reviewSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
No app imports a package's internals: Every exports map exposes .
- Very specific: line 25
Relative specifiers carry no extension: Write from "./domain", never from "./domain.js".
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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
Status: current Last-verified: 2026-09-25
packages/AGENTS.md
Workspace packages shared by the worker and the dashboard. contracts holds
cross-application shapes and constants; conditions evaluates predicates;
costs prices provider usage; harness owns model policy and built-in
compatibility profiles; prompts owns prompt composition; skills
owns browser-safe product skill contracts and validation;
workflow-graph owns the pure rules of a workflow definition; and
agent-visibility owns the record of what one send gave a model and the
clarification rounds, and never composes a prompt. These pure
packages may import another shared package only through its public entry point
and never application infrastructure. ADR-001 owns the tiers.
The rules that bind
- Source entry, no build.
main,typesandexports["."]all point atindex.ts. There is nodist, nobuildscript and nobuild:sharedanywhere. Nitro and Next inline the source into their own bundles. - No app imports a package's internals. Every
exportsmap exposes.only. Import a package through its public entry and add a re-export to itsindex.tsinstead of deepening an import. - Relative specifiers carry no extension. Write
from "./domain", neverfrom "./domain.js". Nitro tolerates the.jsform, webpack does not, and the failure appears only in the dashboard build. - Inlining relies on pnpm's symlinked node-linker. The Vercel tracer
bundles a workspace
.tsentry because its realpath resolves inside the workspace. A hoisted or isolated node-linker would send it looking for adistentry, and there is none. - Each package typechecks itself. A package with no scripts drops silently
out of
pnpm -r typecheck, so each keeps atypecheckscript and a stricttsconfig.json. - The test scripts name the packages they run, and they reach past
packages/intointegrations/.pnpm run test:packagesrunstestin every shared package and every integration that owns atestscript (the template,@integrations/example, included), andpnpm run test:packages:zod4runstest:zod4in the packages that carry a runtime zod dependency, every provider integration, the template, the SDK and the registry, whose conformance sweep parses every integration's schemas and so has to run under the zod production loads. The--filterlists in the rootpackage.jsonare the one list; that sweep also covers the fixtures underintegrations/_fixtures/, which have no suite of their own. Both scripts name packages with--filter, because--if-presentover./packages/*reports a package that owns no such script as passed.conditionshas notestscript at all and no suite in either run; that is a gap, not a decision recorded here. The test inscripts/ci/verify-changed.test.tsholds each named list equal to the packages that own the script, across both roots, so adding a package to a run is a deliberate edit and leaving one out is a failing test rather than a silent opt out. - Shared dependency versions live in the root catalog. Anything two
projects declare goes on
catalog:, enforced byscripts/gates/check-deps-consistency.mjs.
Area rules
.claude/rules/ carries what is true of one package only, loaded when a
matching file is read: workflow-graph (what may live in the package, its
suites and gates), contracts-requests (request body schemas),
zod-bundle (the zod the worker bundle really runs),
worker-settings (the settings registry),
agent-visibility (the record of what one send gave a model),
sandbox-agents (packages/prompts), and in integrations/, adapters,
memory (mem0) and zod-bundle.
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.

