Agent Command Center

A local-first command center for supervising AI-assisted tasks, approvals, execution logs, validations, receipts, and archived work records.

Agent Command Center is built around one workflow:

Task -> Plan -> Approval -> ExecutionLog -> Validation -> Receipt

AI coding agents can be powerful, but their work can become hard to supervise. This project makes agent work more transparent by requiring reviewable plans, explicit approvals, execution logs, validation notes, and receipts before work is treated as complete.

The first public release is demo-only by default. It uses fictional sample data, requires no API key for the demo, and keeps real execution disabled by default. Codex, OpenClaw, QClaw, and GPT are represented as configurable placeholder integrations, not as automatically connected executors.

Why It Exists

AI agents can move quickly, but the surrounding workflow often becomes hard to audit: what was requested, what plan was approved, what changed, what evidence exists, and what should happen next.

Agent Command Center is a local-first supervision layer for that workflow. It is not a chat shell and it is not a hosted automation platform. It is a reviewable local workbench for task intake, approval, queue tracking, validation notes, and receipts.

The goal is to make agent-assisted work easier to inspect:

• Every task should have a reviewable plan.
• Risky actions should require approval.
• Execution should leave a log.
• Results should be validated.
• Completed work should produce a receipt.

Core Features

• Lightweight home screen for the current work focus.
• Side panel for goals, recent records, drafts, logs, and approvals.
• Command bar for registering demo tasks and opening local demo records.
• Manual approval before a task enters the local queue.
• Queue status updates with manual result and validation notes.
• Receipt records for finished demo work.
• Sample local ledger stored in examples/sample-data/.
• Safe runtime demo state written to data/demo-state.json, ignored by Git.

Architecture Preview

User Command
  -> Task
  -> Plan
  -> Approval
  -> ExecutionLog
  -> Validation
  -> Receipt
  -> Local Ledger

The React front end reads state from the demo server in server.mjs. During Vite development it can also fall back to public/data/. The demo server reads fictional records from examples/sample-data/ and writes local runtime state to ignored files under data/.

Safety First

• No real agent execution by default.
• No API key required for the demo.
• Local file access should be configured with allowlists.
• Demo data only: no private runtime logs, receipts, task cards, or uploads.
• External integrations are configurable placeholders and remain off unless explicitly implemented and enabled.

All real capabilities are off by default:

ENABLE_REAL_EXECUTION=false
ENABLE_OPENCLAW=false
ENABLE_QCLAW=false
ENABLE_CODEX=false
ENABLE_FILE_UPLOAD=false
ENABLE_DAILY_SCAN=false

Do not commit real runtime data, private logs, receipts, task cards, uploaded files, credentials, or local machine paths.

Task Flow

Task -> Plan -> Approval -> ExecutionLog -> Validation -> Receipt

In demo mode, this flow is simulated locally:

1. Create or open a task.
2. Review a generated demo plan.
3. Approve manually.
4. Track the item in a local queue.
5. Record manual validation notes.
6. Keep a receipt.

No real executor runs as part of this flow.

Quick Start

npm install
npm run dev

Open the local URL printed by Vite. The app falls back to public/data/ during Vite development.

To run the demo server after building:

npm run build
npm run start

Then open http://127.0.0.1:5179.

Demo Data

The public demo data lives in:

examples/sample-data/
public/data/

Both folders contain fictional records only. They are meant to demonstrate the home screen, side panel, task detail drawer, approval queue, logs, and receipts without exposing private work.

Runtime changes created by the demo server are written to:

data/demo-state.json

The data/ folder is ignored by Git.

Screenshots

Home

Sidebar

Task Detail

Command Bar

Integrations

The repository includes configurable placeholders for:

• Codex Demo
• OpenClaw Demo
• QClaw Demo
• GPT Link Demo
• Manual Demo

These are not automatically connected to real tools. The first public version only records demo plans, approvals, queue items, and receipts. Future integrations should be opt-in, configured through environment variables, protected by allowlists, and gated by manual approval.

Useful Commands

npm run smoke
npm run build
npm run dev
npm run start

npm run smoke checks required project files, sample data folders, and obvious sensitive keywords while skipping node_modules, dist, and .git.

Project Structure

agent-command-center/
  src/
  public/data/
  examples/sample-data/
  docs/
  screenshots/
  scripts/smoke-check.mjs
  server.mjs

Documentation

• docs/architecture.md
• docs/task-flow.md
• docs/security.md
• docs/integrations.md
• docs/good-first-issues.md
• screenshots/README.md

Roadmap

• Phase 1: public demo with fictional data and no real execution.
• Phase 2: configurable local ledger with explicit file allowlists.
• Phase 3: optional agent integrations that remain off by default.

See ROADMAP.md for more detail.

Release Notes

See RELEASE_NOTES.md.

License

Apache-2.0