Forge and Skill Recorder turn a repeatable job into an owned, reviewable asset.
Make something that should exist
Reuse first. Try safely. Forge only what is missing. 180 agents live in this tenant · 8 skills on the shelf.
Does this already exist?
Search your owned skills, forged agents and the open community before spending a credit. Reuse costs nothing.
Couldn’t search just now — try again.
You’ve reached today’s search limit. Try again tomorrow — or say it below and I’ll still check for duplicates before spending anything.
Where Aristo searches, and what could not be checked
Every source shows its own state — not searched yet, searched, partial, unavailable or not connected. A source that could not answer is never counted as “nothing found”, and while one of your own sources is unavailable I will not spend a credit on the guess that this does not already exist.
Raising a real request needs a ticketing system connected to your tenant. Nothing is connected yet, so this stays off rather than opening a form that goes nowhere.
Try it first — nothing saved, nothing counted
A safe stage-zero. Run a prompt, dry-run a skill on content you provide, or chat with an agent. Aristo uses the configured AI runner when it is available; a skill dry-run falls back to a clearly labelled structural readiness check when it is not. Either way, this stage writes nothing and counts nothing. When one earns it, Graduate to Forge to own it.
This is an Aristo dry-run on the content you provide — not Copilot-in-Excel.
Forge is where it becomes owned, spent-on, and counted.
Say it
Picked up where you left off — you started this , and it was still here.
Your shaped agent and your answers are back. The security attestation is not: a signature has to be fresh, so press Attest again before you forge.
What should exist that doesn’t?
Couldn’t shape it — try again.
The sandbox has done its share of shaping for today — try again tomorrow. Nothing was spent, and nothing was lost.
What this will cost, before it costs it
already lives in this tenant — forged once, owned by your people. Use what exists before a credit is spent.
Shape · Forge · Own happen right here — the package lands below, counted.
Shape it
For:
Security & compliance — a quick self-attestation, required before you build.
Nothing is pre-selected. Answer all six — an unanswered question is not a safe answer, and the build is not signed until every one of them is.
Couldn’t forge — try again.
The sandbox has done its share of forging for today — try again tomorrow. Nothing was spent, and your shaped spec is still here.
This is your own self-attestation — recorded on the package and your governance registry, not an Aristo-verified compliance check.
Own it
is built — a complete agent package you own.
Upload it yourself: Teams → Apps → Manage your apps → Upload an app. $0/mo added on licensed seats.
Make a SKILL — build once, reuse in Copilot Cowork
Say the need. Aristo shapes a real SKILL.md (the open standard), assays it against a 10-mark quality rubric, and if it clears the bar it lands on your org Skill Shelf — teammates can reuse it in Copilot Cowork with no Aristo credit re-burn.
What this will cost, before it costs it
Couldn’t shape it — try again.
is already on your shelf. Use it free — or forge a variant ().
Couldn’t forge it just now — nothing was lost and nothing extra was spent. Your shaped skill is still here; press Forge again.
Assay score
Scored so your org can trust it unread.
On the Shelf. Built once. Every teammate who runs it now spends zero credits.
Get the Cowork-ready folder (.zip) Download the SKILL.md
The .zip contains <skill-name>/SKILL.md. In Copilot Cowork, use Customize › Skills › Add › Upload skill, or place the unzipped folder under OneDrive Documents › Cowork › skills; Cowork discovers it at the start of the next session. The .md is the portable open format.
Prepare for organization-wide distribution →
The shelf ZIP is a personal Cowork skill, not a Microsoft 365 app package. Organization-wide distribution requires an admin-reviewed plugin/app package and your tenant’s approved assignment flow. Aristo hands this off; it never claims the skill is published.
Scored against Aristo’s 10-mark rubric · confirm with one real run in Copilot Cowork before you rely on it.
Two fixes to clear the bar:
Bring a recorded skill
Don’t want to write it or prompt it? Show it. Record yourself doing the task once with Microsoft’s Skill Recorder, then bring the SKILL.md back here — Aristo checks it, scores it on the same 10-mark rubric, and shelves it under your name, labelled recorded.
The recording happens on your own machine — the recorder captures your screen locally.
Its Analyze step is the part to know about: that one sends screen frames, window titles and URLs, and your spoken narration to GitHub’s cloud to turn the demonstration into a skill.
So never record credentials, secrets, or a screen with tenant-sensitive data on it. Record the job, not the sign-in.
Aristo never receives or stores a recording. The only thing that reaches us is the finished SKILL.md you choose to bring — that is true by architecture, not by policy.
A skill file is read by Copilot for everyone it reaches, so an imported one is checked before it is scored: size, a real trigger, and anything in it that speaks past the task. Whatever the file says it may run is shown to you word for word — never quietly granted, never quietly removed.
Own what exists
180 agents already live in this tenant.
8 skills on the shelf — reused 41 times, credits never re-burned.
Spin a flow
Routines you repeat become real Power Automate flows — described, defined, and ready to review & turn on.
Every automatable cookbook carries its own “Build this in Power Automate”.
From a forged package to Prod
Every forged agent enters the pipeline at the first stage. Promotion is governed: the automated assay runs on the way into Security & Compliance, and Prod needs both a passing assay and a fresh human approval. This is Aristo-internal governance metadata — it tracks and gates your stages; it does not move the agent between your Power Platform environments.
BDE Business Dev Environment
Where a group of business people shape the agent together, before anyone spends a credit on it.Nothing is gated here — anyone working on it can move it forward.
ODAT Org dev acceptance testing
The people who did not build it try it, and the business decides whether it does the job.Moving it on runs the automated assay over the built package.
S&C Security & Compliance
The automated assay runs the moment it arrives, so a person only rules on what the machine could not clear.It leaves here only with a passing assay AND a fresh, unexpired human approval.
Prod Production
A sealed package your organization owns, plus a governance record of how it got here.This tracks your stages; it does not deploy the agent into your environments — Aristo holds no write scopes for that.
Built once, reused without re-burn
A forged agent is a package you own; a shipped skill lands on your org Skill Shelf and teammates can import it into Copilot Cowork without re-burning Aristo credits. Organization-wide distribution requires a separately packaged and admin-reviewed Microsoft 365 plugin/app. Aristo hands that work off and never claims it published for you.