demo-data-simulator: synthetic demo data for workflow apps

Generate deterministic synthetic demo data, seed data, fixtures, and dashboard data for SaaS and business workflow apps.

demo-data-simulator lets Codex, Claude Code, or another local agent infer how your app works, then turns that inference into deterministic CSV, JSONL, and optional SQL seed data.

The split is deliberate:

• the CLI collects bounded, redacted project evidence
• your agent infers the entities, relationships, events, scenarios, and metrics
• local deterministic code validates the spec and generates repeatable data

Use it when you need believable demo, seed, fixture, or dashboard data for a workflow/SaaS app without hand-writing every table and event stream.

Requires Node.js 20+.

Best For

• SaaS product demos that need coherent sample customers, accounts, work items, and events
• local seed data for development and QA
• analytics dashboards that need repeatable CSV and JSONL inputs
• AI-agent evaluation fixtures with realistic workflow state
• prototype environments where fake rows need to behave like an operating business

What Makes It Different

Most fake-data tools generate fields. demo-data-simulator generates workflow data: linked entities, lifecycle events, daily metrics, and a reproducibility manifest that agree with each other for the same spec and seed.

Try It In 60 Seconds

No install needed:

npx demo-data-simulator init --project .
npx demo-data-simulator validate --spec simulator.spec.json
npx demo-data-simulator generate --spec simulator.spec.json --seed 42 --out demo-data

Or install it globally:

npm install -g demo-data-simulator
dds init --project .
dds validate --spec simulator.spec.json
dds generate --spec simulator.spec.json --seed 42 --out demo-data

Start from a built-in scenario pack when you want a reviewed demo shape without repo inference:

npx demo-data-simulator init --pack field-service --project .
npx demo-data-simulator generate --spec simulator.spec.json --seed field-service-demo --out demo-data
npx demo-data-simulator proof --spec simulator.spec.json --data demo-data --out demo-data/proof.md

Built-In Scenario Packs

Use a pack when you want coherent synthetic data for a common workflow without waiting for agent inference.

Pack ID	Best For	Includes
field-service	service operations demos and dispatch dashboards	customers, technicians, work orders, scheduling/completion events, overdue and capacity-pressure scenarios
sales-pipeline	CRM, revenue operations, and sales dashboard demos	accounts, sales reps, opportunities, stage-change events, stalled deals, expansion signals, and close outcomes
recruiting-pipeline	ATS, hiring, and people-ops workflow demos	candidates, jobs, recruiters, applications, interview events, offer acceptance, declined offers, and stalled loops

Recommended first-run commands:

npx demo-data-simulator init --pack sales-pipeline --project .
npx demo-data-simulator lint --spec simulator.spec.json
npx demo-data-simulator generate --spec simulator.spec.json --seed sales-demo --out demo-data
npx demo-data-simulator proof --spec simulator.spec.json --data demo-data --out demo-data/proof.md

npx demo-data-simulator init --pack recruiting-pipeline --project .
npx demo-data-simulator generate --spec simulator.spec.json --seed recruiting-demo --out demo-data
npx demo-data-simulator proof --spec simulator.spec.json --data demo-data --out demo-data/proof.md

Outputs:

• demo-data/entities/*.csv
• demo-data/events.jsonl
• demo-data/metrics_daily.csv
• demo-data/seed.sql
• demo-data/manifest.json
• demo-data/proof.md when proof generation is requested

For a quick visual proof, see the synthetic field-service dashboard example in examples/field-service/dashboard/. It reads generated events.jsonl and metrics_daily.csv files, and stays entirely local.

To inspect or review packs before generation:

npx demo-data-simulator pack list
npx demo-data-simulator pack export --pack field-service --out field-service.simulator.spec.json
npx demo-data-simulator lint --spec field-service.simulator.spec.json

For maintainer pack authoring gates, including proof JSON diffing, see docs/scenario-pack-authoring-quality-gates.md.

Public Repo Guide

• User guide and quick start: this README
• Field-service example app: examples/field-service/
• Static dashboard example: examples/field-service/dashboard/
• Maintainer docs index: docs/README.md
• Release runbook: docs/release-runbook.md

Infer From A Repo

Run the CLI inside a product repo and let your existing coding agent infer the simulator plan.

The CLI supports Codex, Claude Code, manual mode, and a normalized command adapter.

npx demo-data-simulator infer --agent codex --project . --profile fast --accept-generated
npx demo-data-simulator validate --spec simulator.spec.json
npx demo-data-simulator generate --spec simulator.spec.json --seed 42 --out demo-data

Codex uses codex exec with a JSON Schema response contract.

Claude Code uses print mode with --output-format json and --json-schema; dds doctor --agent claude currently verifies the binary is present, not that auth and structured-output mode are fully ready.

npx demo-data-simulator infer --agent claude --project . --accept-generated

Manual mode is available when you want to write or edit the spec yourself:

npx demo-data-simulator infer --agent none --project .

That writes NEEDS_DECISION.md with the next manual step.

Any other local agent command can be used if it reads the prompt from stdin and prints the strict inference envelope as JSON:

npx demo-data-simulator infer \
  --agent command \
  --project . \
  --agent-cmd your-agent \
  --agent-arg --json \
  --accept-generated

For custom agent CLIs, wire their non-interactive mode through --agent-cmd and repeat --agent-arg for each argument.

What It Generates

The generated output is intentionally boring and useful:

• linked entity tables in entities/*.csv
• event history in events.jsonl
• daily metrics in metrics_daily.csv
• optional SQL inserts in seed.sql
• a reproducibility manifest in manifest.json
• optional proof reports in Markdown and JSON

The generated rows are deterministic for the same spec and seed.

Use CSV when you want spreadsheet-friendly entity tables or fixtures, JSONL when you want append-style event streams for dashboards and agent evaluations, and SQL when you want to seed a relational database directly. SQL inserts are written in dependency order: referenced entity rows first, dependent entity rows next, then events and daily metrics.

Why Not Faker?

Faker makes fields. This makes workflow data.

For example, faker can create a customer name. demo-data-simulator is meant to create customers, work orders, technicians, assignment events, completion events, exceptions, and metrics that agree with each other.

That matters when you are building:

• SaaS product demos
• local seed data
• analytics dashboards
• AI-agent evaluation fixtures
• sales or prototype environments

Why Not Just Prompt An Agent?

Agents are good at inference. They are less reliable as the whole runtime.

This package keeps the agent on the part it is good at: reading bounded evidence and drafting the simulator spec. The CLI handles the parts that should be boring and repeatable:

• evidence collection and source prioritization
• secret redaction
• a strict JSON inference contract
• spec validation
• deterministic seeded generation
• CSV/JSONL/SQL writers
• CI-friendly commands

That means the same inferred spec can be reviewed, committed, regenerated, and tested without asking an LLM to recreate rows every time.

Reproducibility check:

dds generate --spec simulator.spec.json --seed 42 --out demo-data-a
dds generate --spec simulator.spec.json --seed 42 --out demo-data-b
diff -ru demo-data-a demo-data-b

CLI Core, Skill Layer

This can pair well with agent skill packs. A skill can teach Codex or Claude when to call dds, how to review the generated spec, and how to improve it for a repo.

The CLI remains the durable engine. It gives every agent the same evidence boundary, schema, validator, generator, and output format.

This package includes an agent skill at skills/demo-data-simulator/SKILL.md. Agent runtimes that support skill folders can copy or reference that directory so the agent knows when to use dds, how to review simulator.spec.json, and how to generate data safely.

Pre-Publish Local Pack Smoke Test

Before any public npm release, verify the package from a local tarball:

npm run smoke:pack

The smoke test runs npm pack, installs the generated tarball into a clean temporary project, then verifies the packaged dds binary can print help, validate a sample spec, and generate sample CSV/JSONL/manifest output. It does not publish to npm, configure trusted publishing, or change account settings.

For the current local approval evidence and remaining npm gates, see docs/pre-publish-proof-bundle.md. For the automated tag release process and trusted publishing setup, see docs/release-runbook.md.

Commands

  dds doctor --agent auto
  dds init --project .
  dds init --pack field-service --project .
  dds pack list
  dds pack export --pack field-service --out field-service.simulator.spec.json
  dds infer --agent codex --project .
  dds infer --agent codex --project . --profile fast
  dds infer --agent claude --project .
  dds infer --agent command --agent-cmd <bin> --agent-arg <arg> --project .
  dds validate --spec simulator.spec.json
  dds lint --spec simulator.spec.json
  dds generate --spec simulator.spec.json --seed 42 --out demo-data
  dds proof --spec simulator.spec.json --data demo-data --out demo-data/proof.md
  dds proof diff --baseline proof-before.json --candidate proof-after.json
  dds explain --spec simulator.spec.json

Safety Model

The CLI collects a bounded evidence bundle before invoking an agent:

• skips dependency/build directories, binary files, dot-env files, credentials, tokens, and key files
• only reads allowlisted text/source extensions
• applies simple secret redaction before building the prompt
• writes .demo-data-simulator/evidence-manifest.json
• treats repo contents as untrusted evidence, not instructions
• ranks candidate files before spending the evidence budget, prioritizing source, schema, model, API, workflow, test, and fixture files over low-signal config/docs

Evidence profiles:

• --profile fast: smaller first-run bundle for larger repos
• --profile balanced: default
• --profile wide: larger bundle for deeper inference

The built-in Codex preset is run with a read-only sandbox. Custom --agent-cmd commands are user-controlled, so review those commands the same way you would review any local script. The generated spec is written to .demo-data-simulator/simulator.spec.generated.json; simulator.spec.json is user-owned.

Spec

MVP specs are JSON only and use schemaVersion: "simulator.v1". A spec defines entities, fields, relationships, events, scenarios, metrics, and outputs. Set outputs.formats to any mix of csv, jsonl, sql, and manifest.

See examples/specs/field-service.simulator.spec.json.

When Not To Use It

This is not the best tool for:

• one-off random names or addresses
• production anonymization
• load testing with millions of rows
• domains with no workflow, state changes, or relationships

For those, a faker library, anonymization pipeline, or load-test generator is probably a better fit.