Anlyx

See what your AI Agent understood about your project.

Anlyx reads an AI Agent-authored anlyx.project.json and opens a local visual workspace for pages, features, requests, architecture, evidence, and gaps.

Quick Start

Install Anlyx in the project you want to inspect:

npm install -D anlyx

Ask your AI Agent to create anlyx.project.json at the project root. Then run:

npx anlyx dev

The viewer opens locally, usually at:

http://localhost:4777

What It Shows

• Overview: project summary, main areas, stack, and unresolved gaps.
• Pages: page stories, user actions, linked requests, and flow details.
• Map: authored architecture nodes and edges as a visual connection map.
• Capabilities: product features with related pages, requests, data, evidence, and gaps.
• JSON: the raw Project JSON, counts, validation state, and source issues.

Anlyx does not invent missing pages, requests, architecture edges, evidence, or timing. If the Agent could not prove something, it should stay inferred, missing, unresolved, or unknown.

Agent Prompt

Paste this into Codex, Claude Code, Cursor, or another coding Agent:

Set up Anlyx inside this local repository so the Project JSON viewer can run.

Read the Anlyx guidance docs first. If this repository does not include them, use:
https://suhannoh.github.io/anlyx/docs

If this is a monorepo or split frontend/backend project, keep the
user-confirmed repository root as the analysis root. If that root has no
package.json, install Anlyx in the appropriate child Node package only so the
CLI is available, but write anlyx.project.json at the analysis root. Run the CLI
from the analysis root with the child binary path when needed, for example
./frontend/node_modules/.bin/anlyx.

Create or update anlyx.project.json with schemaVersion "0.2.0".
Use only local files, routes, pages, components, API handlers, services, schemas,
config, and docs you actually inspect.
Do not invent pages, requests, flows, architecture edges, timing, or evidence.
Do not include secrets, production records, or raw personal data.

For large projects, write an early valid checkpoint JSON before exhaustive
coverage so schema and viewer issues are caught quickly, but do not treat that
checkpoint as final setup. Build an inventory of known pages, routes, requests,
flows, capabilities, architecture nodes/edges, and evidence, then continue
enriching anlyx.project.json until the inventory is covered. If something cannot
be proven or inspected locally, record it as unknown, not-proven, missing,
unresolved, or unresolvedGaps. Only call setup complete when known project
surfaces are authored or explicitly named as unresolved.
Before the final response, reconcile the discovered inventory with
anlyx.project.json. If a known route, page, request, backend node, or major flow
is missing, add it or record why it remains unresolved.

Do not limit AI / Agent Tools to repository lock files only. If the user tells
you this project used tools such as Superpowers, Notion, Slack, ImnotAI, MCP
servers, plugins, or skills, add them to project.metadata.agentTools with
provenance "user-provided". Mark a tool "project-configured" only when
repository evidence such as a lock file, MCP config, plugin manifest, package
dependency, or workflow doc proves it. Do not inspect or guess private Codex App,
Claude App, Slack, Notion, browser, or OS app history.

If you author project.icon, choose the logo or mark the product actively shows
to users. Prefer a project-relative path, URL, or SVG when possible. Use
data-uri only for small self-contained assets when no path or URL is available.

Author the Map as readable dependency paths, not as a framework or route
inventory. Pages should carry detailed page/request coverage. Map nodes should
usually follow entry flow -> request/API -> controller/handler ->
service/policy -> repository/data/external. Avoid broad product-entry nodes such
as Browser, Next.js, React, Spring Boot, Nginx, Docker, Public routes, Admin
routes, or All controllers unless they are specific inspected infrastructure
nodes. Split high-degree hubs by domain, endpoint family, or representative
entry flow.

Run:
1. npm install -D anlyx
2. npx anlyx dev

Verify Overview, Pages, Map, Capabilities, and JSON open without broken screens.
Do not write a long analysis report. Only state whether setup is ready, the
command to run, and the files changed.

Project JSON

The default input is:

anlyx.project.json

Minimal shape:

{
  "schemaVersion": "0.2.0",
  "project": {
    "id": "example-project",
    "name": "Example Project",
    "generatedBy": {
      "kind": "agent",
      "name": "Codex"
    }
  },
  "areas": [],
  "pages": [],
  "features": [],
  "requests": [],
  "flows": [],
  "architecture": {
    "nodes": [],
    "edges": []
  },
  "evidence": [],
  "measurements": [],
  "dictionary": {
    "defaultLanguage": "en",
    "terms": []
  }
}

Full docs:

• https://suhannoh.github.io/anlyx/docs
• https://github.com/suhannoh/anlyx