Skill · Testing
api-tests
+340 todayv3
Your own blocks and pages
Your own start page and pages, built in the chat or from templates.
IDEFrameworkInternal PlatformCommunity
Share Skills, Agents and Workflows. Every AI in your project uses them.
You write rules, Skills and Agents once. Trinity turns them into the files for Claude Code, Copilot, Cursor and co., shows you live what your agent is doing, and shares what works with your team.
What others have built, you adopt with one click. Your AI first checks where it belongs in your project, and nothing is written until you say OK.
Works with
Claude CodeGitHub CopilotCCursorWWindsurfOOpenClaw+any other AI
Favourites
DesktopDownloadsDocumentsProjects‹ › Downloads
⌕ Search
NameDate modifiedSizeKind
trinity-workbenchToday, 09:38--Folder
angebot-aurora.pdfYesterday, 16:12412 KBPDF document
notizen.txt31.10.2026, 11:052 KBText
bildschirmfoto.png30.10.2026, 08:471,2 MBPNG image
Favourites
DesktopDownloadsDocumentsProjects‹ › aurora-app
⌕ Search
NameDate modifiedSizeKind
.github12.10.2026, 10:02--Folder
docs28.10.2026, 14:31--Folder
srcYesterday, 17:54--Folder
trinity-workbenchToday, 09:41--Folder
.gitignore12.10.2026, 10:021 KBText
package.jsonYesterday, 17:543 KBJSON
README.md28.10.2026, 14:315 KBMarkdown
“trinity-workbench” is being copied to “aurora-app”About 2 seconds left
Trinity
Trinity · aurora-app
trinity-workbench+
Trinity is your AI universe.
One folder into your project. For every AI.
What is Trinity?
No more five AI instruction files that contradict each other.
Everything your AI needs for the project, in one place.Today it’s spread across CLAUDE.md, the Copilot instructions, Cursor rules, notes and the wiki. Trinity collects it in your project and gives every AI exactly the file it expects.
CLAUDE.mdinstructions.md.cursorrulesNotesWikiTicketsClaude CodeCopilotCursorWindsurfOpenClawyour projectResult
CLAUDE.mdinstructions.md.cursorrulesNotesWikiTicketsTrinityClaude CodeCopilotCursorWindsurfOpenClawyour projectResult
What changesThree things you no longer have to worry about.
No more copying the same rule into CLAUDE.md, the Copilot file and the Cursor file.
Write once, every AI follows it.
You create the rule once in the Brain. Trinity writes it where Claude Code, Copilot, Cursor, Windsurf and OpenClaw look.
brain/regelwerke/api.md→ 5 AIs
Brainregelwerke/api.mdClaude CodeCLAUDE.mdCopilot.github/copilot-instructions.mdCursor.cursor/rules/api.mdcWindsurf.windsurf/rules/api.md
No more “So how far along is this?” in the daily.
Tasks run in phases, an agent checks each one.
Brainstorm, Plan, Tasks or your own phases per task type. Once a phase is done, an agent checks in the background whether it’s really met. Only then does the next agent get its assignment.
T-7K3Q · Prices per cityStandard
✓ BrainstormdonePlanyou are here · mode plannerTasksmode implementer
✓ Brainstorm checked by brainstormer · 09:12
BrainstormGoal: a per-minute price of its own for each city.
PlanTable city_prices · priceFor(city, at)
Tasks☐ Create table · ☐ Write priceFor
Never again guessing what the agent did.
Watch live what your agent reads and changes.
Every file read, every rule used, every Skill and every change, as it happens. Afterwards the whole run is in the history and on the task.
Live · T-7K3QClaude Code
09:41:02reads brain/regelwerke/api.md
09:41:05uses Skill pricing-rules
09:42:18changes src/pricing/city.ts+38 −6
09:43:30Check Tests running✓
Just sharedExamples
Share
No more good Skills gathering dust in a single repo.
See a Skill in the feed, adopt it, use it right away.Someone shares a Skill like a post. You adopt it, and from then on every AI in your project can use it. The four steps run on their own, or you click through them.
No more projects where nobody has the overview.
Build your project dashboard in the chat.
Describe what you want to see, or pick a template. The AI turns it into blocks and pages, in your company’s look.
aurora-app · OverviewAurora
BlockFleet 1,284 bikes, 96% ready
BlockOpen tasks 7, 2 of them with agents
PageSprint under Own pagesnew
The flowFrom request to commit.
This is how a task runs with Trinity, from the idea to the commit in the repo.
- TaskBrainstorm, Plan and Tasks end up in one task.md. You plan them in the chat or pull them from a source like SonarQube.Tasks
- Read the BrainThe AI loads your rules, Skills and the state of the project.Brain · Memory
- WorkClaude Code, Copilot or another AI gets going. You watch live.Live · Consoles
- CheckAn agent checks every phase. After every run, Trinity checks the mandatory rules.Phases · Workflows
- ResultLands on the task, along with its history.History
- CommitTrinity commits. You push.Git
Example projectClick through a project with Trinity.
Aurora Mobility is an invented rental service for e-scooters and bikes. Click around freely, or go step by step with “Next”.
Adjustable per projectYou set up every project differently.
Start page, look, rules, Agents, AIs and languages are set per project. No project has to look or work like another.
Overview
Appearance
Your own design
Colours, fonts and shapes per project, matching the company.
Brain
Your own rules
Rule sets, Skills and Agents that apply only in this project.
Workflows
Your own flows
Which Agent goes after which.
AI
Your own AIs
Which AIs work along: Claude Code, Copilot, Cursor or others.
Languages
Your own languages
Interface, work and chat, each in the language you want.
The three sidesTrinity is made of three parts.
IDE
Where you work
The Workbench runs in the browser, right in your project. Chat with drawing board, tasks, live view and a dashboard you build yourselves.
Framework
What your AI reads
Brain and Memory: rules, Skills, Agents and the project status as files in the repo. The same for every AI.
Platform
Where you share
internal-trinity runs on your own server. Teams share their Skills and Puzzles there. A public Community is planned.
Built for the way you workNo new formats, no AI of its own, no push without you.
Open standards
Files your AIs already know
CLAUDE.md, AGENTS.md, copilot-instructions.md, .cursor/rules, Skills, Hooks and MCP. Trinity doesn’t invent its own format for your AI files.
Stays with you
Nothing goes out unless a human clicks
Brain and Memory live in your repo. Trinity only commits; pushing and entering keys is done by a human. internal-trinity runs on your server.
Any AI
No new AI subscription
Trinity brings no AI and no credits of its own. You simply keep working with Claude Code, Copilot or Cursor.
| Type | In the Brain | Claude Code | Copilot | Cursor |
|---|---|---|---|---|
| Instructions | Rule sets | CLAUDE.md | copilot-instructions.md | .cursor/rules/ |
| Skills | skills/ | .claude/skills/ | .github/prompts/ | .cursor/commands/ |
| Hooks | hooks/ | .claude/settings.json | .github/hooks/ | .cursor/hooks.json |
| Tools (MCP) | mcp/ | .mcp.json | .vscode/mcp.json | .cursor/mcp.json |
LocalRuns in your project folder. Brain and Memory live in the repo.
No auto-pushTrinity only commits. Pushing is always your click.
13 languagesFor interface, areas and standards.
Windows and LinuxTested on both.
FAQBriefly explained.
What is Trinity?
A folder called trinity-workbench that you copy into your project. It holds a Brain (rules, Skills, Agents), a Memory (project status, tasks, history) and an interface that runs in the browser. Every AI in the project works with it.
Is Trinity an IDE or an AI of its own?
Neither. Your AI keeps writing the code: Claude Code, Copilot or Cursor. Trinity makes sure every AI knows the same rules and shows you what it’s doing right now.
How is this different from a CLAUDE.md or AGENTS.md?
Those files only apply to certain tools. In Trinity you maintain your rules in one place, and Trinity writes the right file for each tool from them, plus Skills, Hooks and MCP entries. Whatever you write into those files yourself stays put.
What are Brain, Memory, Skill and Puzzle?
Brain is the rules your AI works by (rule sets, modules, Skills, Agents). Memory is the status of your project (project, tasks, assets, history). A Skill is a reusable guide the AI calls for a specific task, like “write tests”. A Puzzle is anything you can share and adopt: Skills, rule sets, Hooks or pages.
What are phases?
The steps of a task, for example Brainstorm, Plan and Tasks. Every task type can have its own phases. Before moving on, an Agent checks in the background whether the current phase is really done.
What happens to my code and my keys?
Everything stays in your repo. Trinity commits, but never pushes. API keys are entered by a human; the AI never gets to see them.
Where do the platforms run?
Each company hosts internal-trinity itself, on its own server. trinity-community, the public platform, is planned.
Can I download Trinity?
Not yet. Trinity isn’t publicly available yet. Feel free to send questions to the contact address at the bottom of the page.
NextTake a closer look at the parts.
IDE
Where you work
Chat, drawing board, tasks with phases, Console Board and live view.
Framework
What your AI reads
Rules, Skills, Agents and the status of your project.
Community
Where you share
Feed, profiles, and adopting Skills with one click.
Trinity isn’t publicly available yet. You’ll find the contact at the bottom of the page.
IDETrinity Workbench
Chat, tasks and Agents in one interface. With the AI of your choice.
The Workbench lives in your project and runs in the browser. Whether Claude Code, Copilot or Cursor works in it is up to you, and you can switch any time. Rules, Skills, tasks and history stay as they are.
Works with
Claude CodeGitHub CopilotCCursorWWindsurfOOpenClaw+any other AI
The filmYour own project pages, built in the chat.
A team describes in the chat what it needs. Trinity builds the page, and everyone in the project has it right away. Every project looks different. Just over a minute, no sound.
With any AISwitch AIs without losing knowledge.
Trinity holds the rules, Skills and project status; the AI is just the engine. You pick it at the bottom of the chat, and each one gets the same knowledge.
No more explaining everything again when you switch AIs.
Different AI, same knowledge.
You pick the AI at the bottom of the chat. Claude Code, Copilot and Cursor then all know the task, its phase, your rules and Skills, because Trinity hands them over.
Chat · T-4127Example
Where are we on T-4127?
Claude CodeGitHub CopilotCursorWindsurf answers
TaskPrices per city, phase Plan
Rulesregelwerke/api.md, preise.md
Skillpricing-rules
NextFinish writing the plan, then Taskssame
AIClaude CodeGitHub CopilotCursorWindsurf▾
No more keeping rules in sync across five files by hand.
Once into the Brain, every AI has the rule.
Trinity writes them where each AI looks first. Via MCP, every AI also reads and writes the project directly, with the same options as the interface.
brain/regelwerke/api.md→ every AI
Brainregelwerke/api.mdClaude CodeCLAUDE.md, .claude/Copilot.github/copilot-instructions.mdCursor.cursor/rules/Windsurf.windsurfrulesOpenClawits instructionsany otherai-assistant.mdMCPtrinity: read and write
No more five terminals where you lose track.
Several Agents side by side on one board.
The Console Board starts any AI with a CLI, such as Claude Code, Copilot, Codex, Gemini, Ollama or OpenClaw. You sort them into groups, put one in charge and hand out sub-agents and handovers. When one is waiting for you, you hear a tone.
Harness normalisationEvery AI gets the same harness.
Rules, Skills, Agents, Workflows, Hooks, tasks and history live in one place, and Trinity hands them to every AI in its own format. So every developer works with the AI they want: Claude Code, Copilot, Cursor or another one. Switching model or provider? No problem, the harness stays the same.
BrainRule sets64 always on, 2 per area
BrainSkills143 adopted and checked
BrainAgents5Model per agent
BrainWorkflows2planner → implementer → reviewer
BrainHooks4new commit → task done
MemoryTasks236 in progress, each in its phase
MemoryAssets41Images, data, drafts
MemoryHistory118Runs with every file read
same harness forClaude CodeGitHub CopilotCursorWindsurfOpenClawany other AI
Example project aurora-app, numbers made up.
Build and checkYou build the agent workflows, Trinity checks the Skills.
No more agents expected to plan, code and review all at once.
Agents work one after another, each with its own job.
Under Brain › Agents › Workflows you set which agent runs after which and which Skill it uses, for example planner, then implementer, then reviewer. A task type starts the workflow, and Live shows where it is right now.
Workflow · tests-fuer-specstarts at: In progress
plannerPlan from the specimplementerSkill playwright · runningreviewerSkill e2e-review
TaskType Feature starts it
Liveimplementer reads specs/login.mdnow
No more copying other people’s Skills blindly into the repo.
New Skills get checked before they go in.
When you adopt a Skill, your AI first looks at where it belongs and whether you already have a similar one. Nothing is written until you say OK. After the run, Trinity checks that everything the Skill is supposed to deliver is there.
Adopt with AI3 parts
newe2e-review → brain/skills/add
similarpricing-rules, yours is newermerge
samelint-basis, already thereskip
Results · T-4127after the run
✓test-report.md with all sections
!review.md, Risks section missingCatch up
In actionThe Workbench in short clips.
Excerpts from the film, with made-up example projects.
Try itClick through the IDE yourself.
The promiseWhat the IDE takes off your plate.
Never again jumping between five windows.
Chat, drawing board, tasks, console and Live sit in one interface.
No more ideas lost in the chat history.
What you plan with the AI becomes a task on the board and a sketch on the drawing board.
No more projects without a dashboard.
Every project gets a home page made of blocks, built yourself or from templates.
No more waiting for a feature you’re missing.
Describe the page in the chat, and the AI builds it right into your interface.
No more setting up every project from scratch.
Templates for task types, blocks and pages give every new project a basic structure right away.
In detailAll the IDE’s features.
- Chat with your AIChat with the AI you already use, right in the project. The chat shows images, links and files as cards.
- Connected to any AIClaude Code, GitHub Copilot, Cursor, Windsurf, OpenClaw and any other via ai-assistant.md or MCP. Under Settings › AI you switch on which ones join in.
- Trinity (Ctrl+K)The assistant knows Brain, Memory and History. Ask about a task or file, and she shows you the spot.
- Drawing board in the chatAsk the AI for a draft, and it draws it with shapes, arrows and text next to the chat. Then you both keep drawing on it.
- Trinity at the boardAsk how something works. Trinity explains it step by step and draws on the drawing board as it goes.
- Tasks from the chatWhatever you discuss, the AI creates right away as a task, with a fixed number, fields and the next step.
- PhasesEvery task shows which phase it’s in. You set your own phases per task type, and an agent checks each one in the background.
- Tasks from sourcesSonarQube, Azure DevOps or any JSON API create tasks in the same format. No more retyping tickets.
- Console BoardSeveral AI sessions side by side, in groups, with a lead, sub-agents and handovers. When one is waiting for you, you hear a tone.
- WorkflowsAgents run one after another, each with its Skill. A task type starts the workflow.
- LiveAs soon as agents are working, you see the workflow: which agent is up, what it reads, uses and changes, and what it’s still missing.
- ResultsThe task says what a Skill is supposed to deliver. If a required result is missing, Trinity tells you at the end of the run.
- OverviewA home page made of blocks. Each block reacts to an event following the pattern Hook → Condition → Action, for example new commit → task done → refresh progress.
- Build pages and blocksDescribe in the chat what you need. The AI turns it into a block that can run as a widget, as its own page in the sidebar and as a window on an event.
- Own tabsWith “+ Tab” you create your own tab, for the team in the repo or just for you. You write your own code and see it right away in the preview. The tab runs in a sandbox and only sees what it’s allowed to.
- TemplatesReady-made task types, blocks and pages, in the project with one click. You share your own templates via the platform.
- ToolsTerminal, editor and web views open in a bar next to the chat. The AI can follow along with what happens there.
- Speaking in consolesPress Tab in the empty input field, speak, send with Tab.
- AppearanceColors, fonts and shapes per project, just for you or for the whole team.
- GraphYour whole project as a network of rules, Skills, Agents and tasks, with every connection between them.
- Own windowsAny page can open as its own window, full screen too.
- AI-readyTrinity checks whether your project has what an AI needs, for example a README with real commands, architecture, a glossary and a guide per AI.
- FocusFocus sessions with music and breaks. At the end you see what you got done today.
- 13 languagesYou set the interface, working language and chat language separately.
In the projectOne folder in the repo, your code stays as it is.
The whole Workbench lives in the trinity-workbench folder. Next to it, Trinity only creates the files your AIs read first, such as CLAUDE.md, .github, .cursor and so on. Trinity doesn’t touch your code.
your-project/ ├─ trinity-workbench/ │ ├─ trinity-app/ # the interface │ ├─ trinity-brain/ # Rules, Skills, Agents │ └─ trinity-memory/ # Tasks, docs, history ├─ CLAUDE.md # for Claude Code ├─ .github/ # for Copilot ├─ .cursor/ # for Cursor ├─ .windsurfrules # for Windsurf ├─ ai-assistant.md # for any other AI └─ src/ … # your code
Next up, the framework: where your rules, Skills and Agents live and how every AI gets them.
FrameworkBrain · Memory
Everything your AI reads before it starts. The same for every AI.
The framework is your rules, Skills, Agents and the state of the project. It all lives as plain files in the repo, and Claude Code, Copilot, Cursor and co. get the same from it.
Brain and MemoryThe Brain says how you work. The Memory says where you are right now.
Two folders in your project. The Brain holds rules, Skills, Agents and Hooks. The Memory hands every new session the project state, so you don’t have to explain everything again.
trinity-brain/Rules, Skills, Agents
rules/Architecture, security, testsconventions/Names, commits, structureskills/“Review code”, “Write tests”agents/Planner, builder, reviewer and their Workflowshooks/what happens on an eventmcp/Tools for every AI
trinity-memory/Tasks, docs, history
project/what it’s about, for whom, which areasdocs/Architecture, operations, APImodules/the parts of your codeproject/tasks/every task with a fixed numberassets/Images, diagrams, specslaeufe/every run, every file read
For every AIWrite it once, every AI sticks to it.
You set up rules, Skills and Agents once. Trinity writes them into CLAUDE.md, the Copilot instructions, the Cursor rules and the files of the other AIs, and shows you what reaches each one.
No more guessing what the AI knows at startup.
See exactly what every AI loads at startup.
For every AI you switch on, Trinity shows what’s in its instructions at startup, what only gets added for matching files and what it doesn’t get at all. Plus a rough token count.
What the AI gets · Claude Code≈ 4,200 tokens
InstructionsCLAUDE.md, Trinity’s parts up to datefully in
Rule setapi.md, sicherheit.mdalways
Rule settests.mdfor *.test.ts
Skills14 with name and descriptionat start
Agents5 with model and toolsat start
MCPtrinity, ticketsystem
No more tools that overwrite your CLAUDE.md.
Trinity only touches its own sections.
In CLAUDE.md and the other AI files, Trinity only maintains the marked blocks. Whatever you write in yourself stays exactly as it is.
CLAUDE.mdby you and by Trinity
# aurora-app We answer in English. Releases on Tuesdays only. <!-- trinity:own-rules (managed by Trinity) --> ## Rule sets - trinity-brain/rules/api.md - trinity-brain/rules/sicherheit.md <!-- /trinity:own-rules --> Questions about payments go to the Checkout team.
your textTrinity’s part
Required means requiredRequired rules are checked after every run.
No more rules the AI just skips.
Required, recommended or optional, you decide.
Every rule and every Skill gets a weight. For a required one, you decide what happens if the AI leaves it out: Trinity reminds it, or it stops and says why. At the end, a script or a review agent checks whether the rule was followed.
Weightingaurora-app
RequiredRule set sicherheitstops
RequiredConvention tests-zuerstChecking agent
RecommendedSkill pricing-rulesreminds
Run T-4127 finishedClaude Code
✓sicherheit read, followed
!tests-zuerst: test came after the codeback to the AI
No more AIs that have to guess how your project is built.
Trinity checks whether your project is AI-ready.
Is there a README with real commands, architecture, a product description, a glossary, conventions and instructions per AI? Whatever’s missing, Trinity creates from a template or hands the job to the AI, but only with your OK.
AI-ready78 %
thereREADME with commands✓
thereInstructions per AI, Trinity’s parts up to date✓
thinProduct description without much textAI job
missingGlossaryfrom template
Try itClick through Brain and Memory.
The promiseWhat the framework takes off your plate.
Never again rules every AI reads differently.
The rule lives once in the Brain. Trinity writes it to the right place for every AI.
No more starting every new session from zero.
The Memory hands every new session the project, docs, tasks and history.
Never again rules that only exist on paper.
Whatever is required, Trinity checks after every run, by script or review agent.
BrainHow your AI works.
- Rule setsHow you work, for example on architecture, security and tests. Every AI gets them as instructions, always or only for matching files.
- ConventionsShort, fixed rules for names, commits and folder structure.
- SkillsAbilities the AI calls, like “Review code” or “Write tests”. Every Skill defines what has to come out at the end.
- Agents and workflowsRoles like planner, developer, tester and reviewer, each with its own model. A Workflow sets who goes after whom.
- HooksWhat happens automatically on an event, like scanning for keys before every commit.
- MCP serversTools every AI can use, for example Trinity itself, a ticket system, wiki or browser.
MemoryThis is where your project is right now.
- ProjectWhat it’s about, who it’s for and which areas there are.
- DocumentationDocs sorted by purpose: architecture, operations and API.
- ModulesThe parts of your code and what they are for.
- TasksEvery task with a fixed number, phases and result.
- AssetsImages, diagrams and specs that docs and tasks point to.
- HistoryWhat has happened so far, including every agent run with every file it read.
How it holds togetherWhat makes it work.
- Required, recommended, optionalEvery rule and every Skill in the Brain gets a weight. The AI has to follow what’s required, and Trinity checks it after every run.
- AI files (PFC)The files every AI reads first, for example CLAUDE.md. Trinity only maintains the marked sections in them; your own text stays.
- What the AI getsOne view shows, per AI, what it loads at startup and what only when needed.
- AI-readyTrinity checks whether the project lacks anything an AI needs, and fills gaps only with your OK.
- Table of contentsEvery folder has a REGISTRY.md. The AI finds what it’s looking for faster and uses fewer tokens.
- Trinity MCPIts own MCP server. Through it, the AI reads and writes tasks, board and Memory, just like you do in the interface.
- LimitsKeys and push stay with a human. That’s built in.
Next up, Internal Platform: how teams in your company share their Skills and Agents.
Internal Platforminternal-trinity
One team builds the Skill, the whole company uses it.
internal-trinity is your company’s own feed for Skills, Agents and Workflows, and you host it yourselves. Teams share what works there, see what gets used most in the company, and adopt it into their projects.
The promiseWhat Internal Platform takes off your plate.
Never again company know-how on someone else’s servers.
You host internal-trinity yourselves. Your Skills and Puzzles stay inside the company.
No more rewriting the same Skill in every team.
A good Skill gets shared once, and every team that needs it adopts it.
No more adopting other people’s Skills blindly.
The AI suggests where each part belongs. Trinity checks the suggestion, and you give the OK.
No more guessing who’s working on what.
You see who’s around and which Agents are running. Only people who turn it on themselves are visible.
Internal Platform · in progressLike a social feed, just for your company’s Skills and Agents.
This is how internal-trinity would look at Aurora Mobility, on their own server. Click through feed, post with comments, adopting with AI, sharing, people, explore and projects.
In detailHow sharing works inside the company.
- Deployed by youEach company runs its own internal-trinity, on its own servers or in its own cloud. What lives there never leaves the company.
- What it is forShare Skills and Puzzles that mustn’t go public. What’s most popular in the company is at the top.
- HomeA feed like on social media. Colleagues share Skills, Workflows and blocks, and you like, comment and adopt them in one click.
- Puzzles and suitesEverything you can share, meaning Skills, rule sets, Hooks and pages, is called a Puzzle in Trinity. Several Puzzles on one topic make a Suite.
- ShareYou submit a Puzzle or a Suite as a version, and a review per topic approves it. All old versions are kept.
- DiscoverA catalogue by topic, sorted by most used or newest. You see how many projects a Puzzle runs in.
- AdoptBy hand or with AI. The AI suggests where each part belongs, Trinity checks the suggestion, and nothing is written until you give the OK.
- PeopleWho’s around right now and which Agents are running, visible only to those who turn it on themselves. You can hand tasks over to a person.
- ProjectsProjects register themselves. The platform only stores a reference, the code stays in your repo.
- OperationsRuns on your side, as Docker with Postgres or as a ZIP.
Next up, the Community: sharing Skills and Agents across company lines.
Communitytrinity-community · planned
Share Skills and Agents with people from other companies.
trinity-community will be the public side of Trinity. People from many companies share Skills, Agents and Workflows there, follow each other and adopt what they need. Your AI checks beforehand where it belongs in your project.
Community · plannedThe public feed for Skills, Agents and Workflows.
The plan: people from many companies share their Puzzles, follow each other and adopt what fits. Here’s how that could look, all content is examples.
trinity-community
Planned · Example
⌕ Skills, Agents, Workflows, people
YOU
Adopted a lot today
last 24 hrsHook · Security
secret-guard
+212 todayv5
Suite · Release
release-paket
+188 todayv3
From people you follow
newest firstAK
Anna K. · NNorthbay Logistics · 3 hrs ago
api-tests 3.0 is here: tests for every new endpoint, now for GraphQL too.
Skill · Testing
api-tests
Spots new endpoints in the diff and writes the tests for them.
runs in 1,204 projectsv3
♥ 812↻ 1,204 adopted✎ 46
FN
Farid N. · HHeron Insurance · 6 hrs ago
Our review Agent now also checks that every change has a task. Saves us hours every week.
Agent · Review
review-lead
Reads diff, task and rules and writes a review with risks.
runs in 486 projectsv6
♥ 301↻ 486 adopted✎ 19
AllTestingSecurityReleaseDataFrontend
Most usedNew
#1
Skill · Testing
api-tests
Tests for every new endpoint.
in 1,204 projectsAK
#2
Hook · Security
secret-guard
Searches for keys before every commit.
in 1,090 projectsBM
#3
Suite · Release
release-paket
Changelog, version number and notes with four Agents.
in 731 projectsMK
#4
Agent · Review
review-lead
Review with risks, task and rules.
in 486 projectsFN
#5
Workflow · Data
import-pruefen
Checks every import against schema and sample data.
in 402 projectsEY
#6
Rule set · Frontend
a11y-basis
Accessibility as rules for every AI.
in 377 projectsLT
Suite · Release · 4 parts
release-paket
MKMara K. · NNorthbay Logistics
Changelog, version number and release notes, written by four Agents. Works with every AI that has Trinity in the project.
◆agents/release-lead.mdAgent
◆agents/notes-writer.mdAgent
◇skills/changelog/SKILL.mdSkill
◇rules/versionierung.mdRule set
v1v2v3 · current
731Projects
42Companies
4.8of 5 in review
MK
Mara K.
Platform at NNorthbay Logistics · shares Skills for releases and reviews
9Puzzles
212projects use them
1,400follow her
Pinned
2 of 9Suite · Release
release-paket
in 731 projectsv3
Skill · Review
pr-summary
in 98 projectsv6
ProfilesYour profile shows where your Skills run.
That’s why sharing pays off. You see how many projects your work runs in and how often others have adopted it.
- ProjectsWhat counts is how many projects your Puzzles run in. Adoptions count, not clicks.
- VersionsAnyone who adopted it sees your new version and decides for themselves whether to update.
- FollowYour followers see your new Puzzles in the “For you” feed.
- CompaniesCompanies get their own profile with their public Suites.
PlannedWhat’s coming to the Community.
- For youA feed with posts from people and companies you follow, plus trends and suggestions.
- ProfilesEvery person shows which Puzzles and Suites they have shared, how many follow them and how often their Puzzles were adopted.
- DiscoverPuzzles from every company, sorted by topic and popularity.
- AdoptSame as internally. The AI suggests, Trinity checks, you give the OK.
Docssoon
The docs are coming soon.
Trinity is still in the works. As soon as it’s available, you’ll find everything here for setting up and working, in every language of the website.
PlannedWhat you’ll find here.
- Guides“You want to … → here’s how”, for example creating a task, building a Hook or sharing a Skill. Every guide shows the steps in the real interface.
- Set upPut Trinity into your project, switch on AIs, host Internal Platform in your own company.
- Brain and MemoryEvery kind of Puzzle with its format, from rule sets and Skills to Agents and Workflows to Hooks and tasks.
- Trinity MCPEvery tool an AI uses to read and write the project.
For companies
Your standards, in every AI and every project.
You set your company standards once in the Brain, and every AI works by them. Everything stays in your repo, and nothing goes out without a human approving it.
The promiseWhat you no longer have to deal with as a company.
No more every team with its own AI rules.
The company standards live in the Brain. Every AI in every project works by them.
Never again code that goes out unchecked.
Trinity never pushes on its own. A push only happens when a human triggers it.
No more API keys in the AI chat.
A human enters keys. The AI never gets to see them.
No more guessing what the Agent did.
Every run is in the history and on the task, with everything that was read, used and changed.
No more being tied to one AI provider.
The Brain stays the same, whether Claude Code, Copilot or another tool is working tomorrow.
No more standards that only one team knows.
You define once what applies everywhere and what only applies to frontend, backend or data.
ControlKeys, push and standards stay with you.
The AI works alongside you, but the critical steps are always done by a human. And you can read up on everything it did afterwards.
No more API keys ending up in the chat history.
A human enters keys, the AI never sees them.
Keys for models and tools live only on the machine where they were entered. The AI learns that a key exists, but never its value.
Settings › Keysthis machine only
ModelANTHROPIC_API_KEY ••••••••set
ToolSONAR_TOKEN ••••••••set
ToolAZURE_DEVOPS_PATmissing
What the AI sees
AI“There’s a key for Anthropic.”
No more code going out without a human looking at it.
Trinity commits, a human pushes.
Agents can work and commit. It only goes into the shared repo once someone on the team clicks “Push”. That’s built in, not just a setting.
Settings › Sharemain → origin
a51f2c0End a ride without signalClaude Code
8dd1b34Rule set payments extendedCopilot
70fbc69T-2224 battery display doneAnna K.
3 commits ready
No more every team with its own AI rules.
Company standards for all projects, also per area.
What applies everywhere, you define once. What only applies to frontend, backend or data, too. New projects pick up the standards during setup, and you see where they run.
Standards · Aurora MobilityExample
alwayssicherheit, commits, tests-zuerst14 projects
Frontenda11y-basis, design-tokens6 projects
Backendapi, payments5 projects
Dataimport-pruefen3 projects
No more guessing what the Agent did.
Every run is in the history.
Which AI, which task, which rules it read and which files it changed. That’s on the task and stays in the repo, also for an audit.
History · T-4127Claude Code · 14 min
readrules/api.md, rules/sicherheit.md
Skillpricing-rules
changedsrc/pricing/city.ts, city.test.ts
Requiredsicherheit followed✓
Commita51f2c0, not pushed
Try itSharing, keys, history and standards in the example project.
Most usedSee which Skills catch on in your company.
A ranking from internal-trinity. It counts how many of the company’s projects a Puzzle runs in.
ControlWhat stays with you.
- Everything in the repoBrain and Memory are ordinary files in your repository.
- No auto-pushTrinity only commits. Pushing happens via Settings › Share, by a person.
- KeysA human enters keys and tokens. They stay on the machine, the AI never sees them.
- TraceableWhat an Agent used and changed is in the history and on the task.
FlexibleFits what you already use.
- Switch AIsClaude Code, Copilot, Cursor, Windsurf or OpenClaw, the Brain stays the same.
- Your own serverInternal Platform runs on your side, as Docker with Postgres or as a ZIP.
- Standards per areaWhat applies everywhere and what only applies to frontend, backend or data.
- 13 languagesFor interface, areas and standards.
SetupIn your project in three steps.
Folder into the project
You copy the trinity-workbench folder into the repo. Everything Trinity needs is inside.Set up
The setup asks for areas and standards and creates the Brain and Memory.Connect your AI
Open your AI in the project. Its instruction file already points to the Brain.
Trinity isn’t publicly available yet. This page shows how it works.