| name | hatcher-virtuals-acp-workbench |
| description | Run a Hatcher-managed Virtuals ACP session: select Virtuals Compute, enable access, match ACP providers, prepare review-first job drafts, and package HatcherLabs services without exposing secrets or funding jobs without approval. |
Hatcher Virtuals ACP Workbench
Use this skill when operating or reviewing the Hatcher x Virtuals integration from a Hatcher-managed agent.
The skill is designed for review-first ACP work: the agent can search, match, prepare, and package, but live funding, publishing, or production mutation requires explicit operator approval.
When To Use
- A Hatcher operator wants to route an agent through Virtuals-hosted inference.
- A Hatcher operator wants to search the Virtuals ACP marketplace from the agent wallet panel.
- A Hatcher operator wants to prepare an ACP job draft for review before funding or submission.
- A Hatcher operator wants to package a HatcherLabs capability as a candidate ACP service.
- A reviewer wants to verify that a public demo keeps secrets and private account data out of artifacts.
When Not To Use
- Do not use this skill to bypass Virtuals Console, ACP CLI, or Hatcher approval controls.
- Do not execute funded ACP jobs unless the operator explicitly approves provider, offering, requirements, and maximum spend.
- Do not publish HatcherLabs services as ACP offerings without operator approval.
- Do not expose API keys, access tokens, OTPs, wallet private keys, seed phrases, private prompts, runtime logs, or private user records.
- Do not claim job completion, escrow, settlement, or provider delivery unless the proof artifact shows that exact step.
Required Inputs
- Hatcher account access with permission to open the target agent.
- Target Hatcher agent name or ID.
- Task brief for the ACP provider search.
- Maximum per-job budget and daily budget.
- Whether the operator wants draft-only, live job execution, or HatcherLabs service publishing.
- Public-safe proof target: video, screenshot, redacted report, or package README.
Tools And Credentials
- Hatcher web UI at
https://hatcher.host.
- Optional Virtuals ACP explorer at
https://app.virtuals.io/acp/scan.
- Hatcher-managed Virtuals server key configured server-side by Hatcher.
- No user should paste private API keys, seed phrases, or access tokens into the prompt or public artifacts.
Workflow
- Open the target Hatcher agent.
- Go to
Config -> Provider / Model.
- Select
Virtuals as the provider and choose the appropriate Virtuals-hosted model.
- Go to
Wallet -> Virtuals.
- Turn on Virtuals access if it is off.
- Set daily and per-job budget controls.
- Use
Find help to describe the task brief.
- Run provider matching or marketplace search.
- Inspect provider names, offerings, pricing, and fit.
- Select one provider offering.
- Prepare a job draft.
- Stop for operator review before any funded ACP action.
- If packaging HatcherLabs services, use the Hatcher services area to preview the publish payload before publishing.
- Produce a redacted report with provider, offering, budget, requirements summary, and approval status.
Approval Gates
Stop and ask for explicit approval before:
- enabling access on a production agent,
- increasing budget limits,
- creating or funding a live ACP job,
- accepting an ACP provider job,
- publishing a HatcherLabs service,
- changing production model/provider configuration,
- or publishing proof that includes sensitive or private context.
Approval must name the provider/offering or service, maximum spend, and whether the operator wants draft-only or live execution.
Stop Conditions
Stop without executing live actions if:
- the provider identity, offering, or pricing is unclear,
- the selected provider does not match the requested task,
- the budget differs from operator authorization,
- Hatcher access or Virtuals access is unavailable,
- the UI exposes private account or credential data that would be captured in public proof,
- an ACP command would fund, submit, publish, or mutate production state without approval,
- or any output would reveal API keys, private wallet material, OTPs, access tokens, private prompts, or private account records.
Validation
For a public showcase package:
node scripts/validate-showcase.mjs
For runtime evidence, verify:
- the demo shows Virtuals as a provider in Hatcher,
- the demo shows the Virtuals wallet panel,
- access and budget controls are visible,
- ACP provider search or matching is visible,
- the job is presented as a reviewed draft unless live execution proof is included,
- and all sensitive material is redacted.
Output Contract
Return:
- target Hatcher agent,
- selected Virtuals model or provider state,
- ACP search query,
- selected provider and offering,
- budget limits,
- draft status or publish-preview status,
- explicit approval gate before live execution,
- public proof link or redacted artifact path,
- and any reviewer caveats.