Digital Employee Quickstart

简体中文

Official case library for Digital Employee. Each case is a portable employee-package.v1alpha1 example with knowledge, Schemas, and offline acceptance fixtures.

Public release boundary

This quickstart is pinned to the public package @fullstack-ai-infra/digital-employee@0.6.0.

Path	Status	What it means here
init, validate, eval	Verified public path	Credential-free package creation and offline contract checks
doctor	Verified public path	Credential-free local readiness check; no authentication and no model call
setup	Released, outside this walkthrough	New in 0.4.0: package setup inside an existing directory; no clean-machine walkthrough record here yet
run	Environment-qualified	Optional one-shot execution with an already configured supported Agent Host
legacy ...	Historical demo/compatibility	standalone-v1 compatibility, not this quickstart's primary experience
deploy	Released, guidance withheld	Retained in public 0.6.0 as a package-bound command: it validates and binds the exact employee package before any deployment effect. Channels: http becomes ready only after authenticated readback; console and dingtalk are preview and never become ready; lark and wecom are unavailable. Runtime is agent-native only; exit codes are 0 ready, 2 pending external action, 1 unsupported or failed. This quickstart gives no deploy instructions or walkthrough until clean-machine acceptance lands in digital-employee#91.

The planned deployment path is tracked in digital-employee#91 and digital-employee-quickstart#2.

Try a case safely

The following path is credential-free and was verified against the exact public release:

git clone https://github.com/bytefolk/digital-employee-quickstart.git
cd digital-employee-quickstart/cases/team-qa

npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee validate . --json
npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee eval . --json

Expected results for team-qa are status: "valid" and then status: "passed" with 3 of 3 fixtures passing.

validate checks the package structure. eval only checks the contract of the repository's offline fixtures. It does not invoke a model, Agent Host, MCP server, or online service, and it does not prove that a digital employee answered these questions correctly.

Report this run

After a success or failure, use the single public feedback entry: comment on digital-employee-quickstart#2. Paste this small, redacted Run Report:

CLI: @fullstack-ai-infra/digital-employee@0.6.0
Case: team-qa
Node / OS:
validate: valid | failed
eval: passed (3/3) | failed
Failure code/output (if any, redacted):

Do not include credentials, account identifiers, private repository names, or local absolute paths.

Cases

Case	Scenario	Description
team-qa	IT team Q&A	Answers from team handbook, on-call rules, permission docs
hr-onboarding	HR onboarding	Guides new hires through processes, benefits, office logistics
ops-approval	Ops approval proposals	Turns requests into structured proposals — never executes
product-faq	Product FAQ	Customer-facing product questions from public docs
sales-qualifier	Sales qualification	Turns a raw inquiry into a structured qualification proposal — never sends, never quotes, requires approval. Optional enterpriseContext lets a company wire in its own CRM data caller-side
sales-employee	General-purpose sales employee	Qualifies a target account, drafts first-touch outreach, extracts commitments/objections/next actions from a meeting transcript, and writes back through a declared crm connector — connector absent means an explicit degraded result, never a fabricated record
support-employee	Customer-service employee	Classifies intent, answers only from cited knowledge-base entries, and escalates below the confidence threshold — with no knowledge base, every request escalates instead of answering from model priors

Run the same credential-free checks against any case:

npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee validate cases/hr-onboarding --json
npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee eval cases/hr-onboarding --json

Showcases

Showcase	Shape	Description
oss-maintainer	1 owner + 3 expert positions	Open-source maintenance business: repo-owner owns the whole business, with issue-researcher, release-engineer, and community-operator as three read-only expert positions
bytefolk	1 organization owner + 7 project leads + 20 specialists	Public-safe adaptation of ByteFolk's real open-source organization, covering seven project teams and 28 offline-verifiable positions
rd-team	1 tech lead + 6 specialists	Public R&D team showcase with read-only packages, handoff contracts, synthetic teaching cases, and deterministic retrieval; no real code-execution loop yet

This category demonstrates the digital-organization workspace target shape: one directory is one business, and one position is one addressable digital employee. The showcase uses the canonical workspace.json, organization.v1alpha1.json, positions/, and context/ layout. It can be inspected with org tree, reconciled with org apply, opened directly in RoleWeave (see the oss-maintainer guide and bytefolk guide for desktop setup and isolation recommendations), and validated with the pinned @fullstack-ai-infra/digital-employee@0.6.0. Interactive chat @position orchestration remains outside this repository's current scope.

Case structure

Every case follows the employee package spec:

<case>/
├── employee.json              # package manifest
├── SKILL.md                   # behavior instructions for the AI engine
├── knowledge/                 # knowledge base (Markdown)
├── schemas/
│   ├── input.schema.json      # input contract
│   └── output.schema.json     # output contract
└── evals/
    └── cases.json             # offline acceptance fixtures

Creating your own cases

Third-party case repositories are welcome. Initialize a package, then validate and evaluate it before attempting a one-shot run with a supported Agent Host:

npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee init my-case --recipe minimal-answer.v1
npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee validate my-case --json
npx --yes --package @fullstack-ai-infra/digital-employee@0.6.0 -- \
  digital-employee eval my-case --json

One-shot execution requires a supported, configured Agent Host. See the v0.6.0 framework documentation for the current engine matrix and run syntax.

Requirements

• Node.js 20 or newer for validation and offline evaluation.
• A supported Agent Host only when performing a one-shot run.

Reference

• Framework CLI documentation
• Employee package specification
• docs/ contains historical deployment drafts. They are not an executable guide for CLI 0.6.0.

License

Apache-2.0. See NOTICE for attribution.