A code-intelligence provider indexes the whole repository. Your coding agent then selects and explains the paths that matter to the reader, producing one clear, evidence-backed HTML map.
Local code index · selective Agent synthesis · standalone HTML
The map starts with the whole system, then lets readers descend only where the detail helps them make sense of the product or implementation.
This is one map, not three examples. Select a level to see what the reader learns as they go deeper.
Start with the product story and one complete path through the system.
The reader gets: the system's purpose, major responsibilities, and primary journey—before implementation detail.
Open one meaningful boundary without expanding the rest of the repository.
The reader gets: module ownership, decisions, interfaces, failures, and outcomes in language suited to their role.
Inspect one node only when its implementation or evidence matters.
This node turns a product rule into an explicit system decision.
backend/app/api/deps.py:25–46The reader gets: purpose, input, output, exact source evidence, and real prompt excerpts where relevant.
There are no rigid “product” and “technical” templates. Your agent translates ordinary language into module boundaries, detail, terminology, prompts, relationships, and progressive disclosure.
Create this for product managers. Emphasize what each module does, user-visible outcomes, and real prompts. Simplify framework and database plumbing.
I’ll present capabilities and user journeys first, merge routine infrastructure, and keep source evidence available behind each node.
Make payment recovery more detailed. Show how the two prompts change the final reply, but hide ordinary API wiring.
I’ll split that module, add prompt-effect relationships, preserve the evidence, revalidate the full path, and replace the HTML only if delivery succeeds.
These are not decorative mockups. Each preview opens a complete standalone HTML generated from a fixed public commit, with a system overview, module views, node evidence, and revisable JSON IR.
Sign-up, sign-in, recovery, and owned items—without making the reader learn container, migration, or generated-client internals.
Best for: product managers, founders, and cross-functional planning.
How declared tasks become queued work, how workers and the scheduler coordinate, and how retries, failures, and results move.
Best for: engineers onboarding to execution and state flow.
The run loop, tools, handoffs, guardrails, sessions, tracing, and a real handoff prompt with exact source evidence.
Best for: product and engineering teams designing agent behavior together.
Install the Codebase Memory repository-intelligence provider, then copy this Skill into your local skills directory. The provider handles exhaustive structural indexing; your agent handles reader-shaped synthesis.
git clone https://github.com/farmerzh47/codebase-system-map.git
cd codebase-system-map
$skillPath = Join-Path $env:USERPROFILE ".codex\skills\codebase-system-map"
New-Item -ItemType Directory -Force $skillPath | Out-Null
Copy-Item ".\codebase-system-map\*" $skillPath -Recurse -Forcegit clone https://github.com/farmerzh47/codebase-system-map.git
cd codebase-system-map
mkdir -p "$HOME/.codex/skills/codebase-system-map"
cp -R codebase-system-map/. "$HOME/.codex/skills/codebase-system-map/"Codebase Memory supplies detailed repository intelligence. Node.js validates and delivers the map; Viz.js, Graphviz, and Expat render its diagrams inside the standalone HTML.
The reference code-intelligence provider for repository indexing, architecture, search, path tracing, and bounded source retrieval.
The JavaScript wrapper and WebAssembly graph renderer embedded in every generated document.
The DOT layout engine compiled into Viz.js WebAssembly, producing the diagrams’ SVG geometry.
An XML parser included transitively in the Viz.js object-code distribution.
Codebase Memory is installed separately and accessed through its Agent Skill or MCP interface. The website is implemented in plain HTML, CSS, and JavaScript. See the complete third-party notices and license texts.
Open source, MIT licensed, cross-platform, and designed to leave you with one file you can actually share.