new-integration
Skill18 starsChanged 58 days agoMITView on GitHub
Add a new integration (plugin) for a third-party provider to AI Workflow, from scaffold to merged pull request, proven with package tests and conformance. Use for "add an integration", "new integration", "new plugin", "connect <provider>", "integrate Linear", "write a Sentry integration", or any request to make a provider usable in workflows.
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- bibektimilsina00/runmycrewtheir own version, first seen May 19, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- bibektimilsina00/runmycrew★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- packages/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
- ai-workflow-reviewSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 50
Write each test so it fails when the rule it names breaks, and see it fail once.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
New integration
The procedure lives in the guide's Start here. This skill walks it in order, says what to show a human at each checkpoint, and names the points where you stop and ask. It restates no rule: when a step says "see", open that section and read it whole.
Read first
- Start here, all of it, then What an integration is and Capabilities.
- The section of the guide for each thing your integration contributes (capability, blocks, webhook, pages), when you reach it.
- The provider's current documentation, through
ctx7, as Before you write a line says. Never from memory. - The shipped package the capability table points at for your case, and
integrations/_template.
Procedure
Record the branch and the full start SHA before editing (root AGENTS.md,
"Evidence"). Run heavy commands one at a time.
- Decide. Fill the decision checklist from the request and the provider's documentation. Checkpoint: post the filled checklist (id, what it contributes, fields with secret or not, settings, connection test call, health checks, webhook yes or no, documentation pages read with dates) and wait for the human's yes. The id and the connection fields are permanent once shipped.
- Scaffold, install, register, check. Steps 1 to 5 of From scaffold to merged. Checkpoint: every command there passes on the untouched scaffold. If one fails before you edited anything, stop and report it: that is a defect in the template or the scaffold, not yours to work around.
- Make it yours, in the order the guide gives: manifest, worker, tests, README. After each file, rerun the package's tests and conformance; rerun typecheck too, except right after the manifest alone: a manifest that declares a capability or a block fails typecheck on its own until the worker matches it, on purpose (the two files hold each other accountable), so treat the manifest and the worker as one checkpoint for typecheck. Write each test so it fails when the rule it names breaks, and see it fail once.
- Review yourself against the review checklist, every box, and run its commands. Checkpoint: paste the commands and their observed outcomes.
- Prove it locally as Prove it says. The production half is an operator's, after the merge: write its steps into the pull request for them, and do not attempt it.
- Open the pull request with what the checklist asks it to say.
Stop and ask a human when
- No capability fits. The provider needs a capability that does not exist
(none of the ones in the guide's table) or the reserved
agent_tools. A new capability is a core design change, not an integration. - The SDK would change. Any edit under
integrations/sdk: a new field on a port, the context or the manifest. It must be additive and recorded in the change log ofdocs/adr/ADR-010-integrations.md, and a human decides it. - Core would change. Any edit under
apps/orpackages/other than the test-mock case the review checklist names for the first provider of a capability. Core never names a provider. - Something shipped would be renamed. An integration id, a block type, a
status variant, a connection field's key,
env,defaultoridentity. That is a migration or a drain, never an edit. - A migration or a database write of any kind. Integrations have no database.
- The scaffold refuses the id because core spells it: the fix may be an allowlist row or another id, and which is right is a human's call.
- You need a credential. A test account's key for a live call that settles what the documentation leaves open. Never use a production key or one of this product's deployments.
- The provider cannot fit the model: it needs an OAuth install flow or a rotating refresh token, or its own events should start workflows (see What an integration cannot do).
- Anything on a deployment: setting variables, connecting, deploying a branch, pressing Test on production. Those are an operator's, after the merge.
Never
Every row of
Things you may not do
binds you. The ones agents break most: "use step" or "use workflow" in
the package, a Node module or global in manifest.ts, reading
process.env, a zod feature zod 4 changed, a webhook test signed by the
function it tests, deploying an unmerged branch, and em or en dashes
anywhere. Do not run the full worker suite locally; CI runs it.
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.

