just-code

agent
Security Audit
Warn
Health Warn
  • License — License: NOASSERTION
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 8 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Just code, nothing else. A community VS Code chat extension on the Claude Agent SDK

README.md

Just Code

Just code, nothing else.

Just Code is a VS Code extension for agentic software development, built on
Anthropic's official Claude Agent SDK.
It provides a chat panel that can read, search, edit, and run code in your workspace —
and a scope policy that keeps the assistant on software engineering, and nothing else.

Independent project. Just Code is not affiliated with, sponsored by, or
endorsed by Anthropic PBC. "Anthropic", "Claude", and "Claude Code" are trademarks
of Anthropic PBC, used here only to describe the underlying SDK and models this
extension consumes.

Why Just Code

The Claude Agent SDK exposes the same agent loop and toolset that powers Anthropic's
own coding tools: a capable general-purpose assistant that happens to be pointed at your
repository. Just Code wraps that loop in a VS Code surface and narrows what it is for.

A scope policy is appended to the SDK's built-in claude_code system prompt on every
request, in every workspace. It confines the assistant to code, debugging, architecture,
and directly-related technical subjects. Personal, social, and sensitive human topics are
out of scope — and so is routing them through code. A request for a function, variable
name, comment, or string literal whose content is out of scope is declined even when
its wrapper is perfectly technical. Anything off-topic gets one answer, and it is always
the same one.

The policy is not a hidden filter. It is roughly ten lines of plain English in a single
file, src/agent/system-prompt.md, so you can read it,
tighten it, or adapt it to your organisation instead of trusting it blindly. It is a
strong instruction to the model rather than a hard technical guarantee, and it is written
to be audited on that basis.

The result is an agent that is a development tool and nothing else — suited to teams and
organisations that want agentic coding assistance without a general-purpose chatbot
inside the editor.

Features

  • Scoped to development — an auditable policy file keeps the assistant on code and
    technical subjects, and declines everything else.
  • Agentic chat panel in the activity bar (and openable in an editor tab), with the
    SDK's full toolset (Read, Write, Edit, MultiEdit, Bash, Grep, Glob, Web, TodoWrite,
    Task/subagents, …).
  • Streaming responses with live text and extended-thinking rendering.
  • Rich tool cards — every tool call is a collapsible card with inputs, results, and
    inline diffs for file edits.
  • Permission prompts — approve or deny each tool use, with "always allow", plus
    permission modes: default, acceptEdits, plan (read-only planning), and
    bypassPermissions.
  • Inline diffs & review — accept/reject individual edits or all at once; rejecting
    reverts the file to its pre-edit snapshot.
  • Editor context — automatically shares your active file and selection; add files or
    selections explicitly via right-click or ⌘⌥K / Ctrl+Alt+K.
  • @-file mentions and /slash commands with autocomplete in the composer.
  • Model picker — switchable mid-session.
  • Usage & cost — per-session token and USD cost indicator.
  • Conversation history — browse and resume past sessions.
  • Checkpoints — rewind file changes made during a turn.
  • Stop / interrupt a running turn at any time.

Keyboard shortcuts

Action macOS Windows/Linux
Open Just Code ⌘Esc Ctrl+Esc
Add selection to chat ⌘⌥K Ctrl+Alt+K
Accept all edits ⌘Enter Ctrl+Enter
New chat (panel focused) ⌘⌥N Ctrl+Alt+N

Requirements

  • VS Code 1.90+

  • Claude Code installed on your machine. Just Code drives Anthropic's own Claude
    Code runtime locally; it does not bundle it (the binary is ~250 MB and platform
    specific). Install it once:

    npm install -g @anthropic-ai/claude-code
    

    The extension auto-detects the executable via PATH, npm global installs, and the
    native installer, following npm launcher scripts to the real binary. If yours lives
    somewhere unusual, set just-code.claudeExecutablePath.

Just Code must run in a trusted workspace. It edits files, executes shell commands,
and starts MCP servers declared in the workspace's .mcp.json — all of which are defined
by the repository you have open.

Authentication

Just Code does not operate a backend. It runs the Claude Agent SDK locally and the
SDK talks to Anthropic directly; your credentials never leave your machine. Two methods
are available (choose via Just Code: Sign In, or the just-code.authMethod
setting):

  • API key — an Anthropic API key from the Claude Console,
    stored in VS Code SecretStorage. Also readable from the ANTHROPIC_API_KEY environment
    variable or the just-code.apiKey setting. This is the authentication method Anthropic
    documents for software built on the Agent SDK.
  • Claude subscription — signs in with your own Claude account through the bundled
    Claude runtime's OAuth flow (claude auth login). In this mode the extension does not
    pass ANTHROPIC_API_KEY to the runtime, so your account login is used rather than
    Console billing.

Before you rely on subscription mode, read Anthropic's
Authentication and credential use
policy and satisfy yourself that your usage fits it. As of this writing that policy
states that developers building products on the Agent SDK "should use API key
authentication", and that Anthropic "does not permit third-party developers to offer
Claude.ai login or to route requests through Free, Pro, or Max plan credentials on
behalf of their users." Just Code routes nothing on anyone's behalf — it invokes
Anthropic's own runtime with your own local credentials — but if you intend to
redistribute Just Code rather than run it yourself, API-key mode is the
unambiguous choice.

Settings

Setting Default Description
just-code.model default Model to use.
just-code.permissionMode default Default permission mode.
just-code.authMethod subscription subscription or apiKey.
just-code.apiKey "" Anthropic API key (prefer the Sign In command).
just-code.claudeExecutablePath "" Path to the claude executable (empty = auto-detect).
just-code.maxTurns 100 Max agent turns per request.
just-code.extendedThinking true Enable extended thinking.
just-code.thinkingBudget 0 Max thinking tokens (0 = model default).
just-code.effort default Reasoning effort level.
just-code.autoModelFallback false Fall back to another model when the primary fails.
just-code.additionalDirectories [] Extra directories the agent may access.
just-code.loadProjectSettings true Load user/project/local settings — enables CLAUDE.md and MCP servers.

MCP

Just Code loads MCP servers from your user configuration (claude mcp add) and from the
workspace .mcp.json. Run /mcp in the chat to see live connection status for each one.
This requires just-code.loadProjectSettings to be on — it is what enables the settings
sources MCP is configured through.

Building from source

Requirements: Node.js 18+ and VS Code 1.90+.

npm install
npm run compile      # builds dist/extension.js and media/webview.js
npm run check-types  # tsc --noEmit

Press F5 in VS Code (or the "Run Extension" launch config) to open an Extension
Development Host with the extension loaded, then open Just Code from the activity bar.

Packaging a VSIX

npm run vsix

The Agent SDK's JavaScript is kept external from the bundle and ships inside the .vsix;
its ~250 MB native runtime is not bundled, so the package is ~4 MB and works on every
platform. The runtime is discovered at load time from your Claude Code installation.

Architecture

src/
  extension.ts            activation, command registration, context wiring
  panel/chatViewProvider  WebviewViewProvider + editor-tab panel; HTML/CSP/bridge
  agent/                  Agent SDK integration (streaming-input query loop)
  agent/system-prompt.md  the scope policy  ← what makes this Just Code
  tools/                  canUseTool permission bridge, diff compute & review
  context/                editor context tracking, @-file / slash completions
  history/                session list & resume via the SDK
packages/core/src/        @just-code/core: typed message contract (protocol.ts)  ← frozen, shared
webview-ui/src/           vanilla-TS chat UI (bundled to media/webview.js)
media/                    webview.css, icons, bundled webview.js
docs/SDK-NOTES.md         verified notes on the Agent SDK surface

The extension host holds one long-lived query() per conversation in streaming input
mode
, translating the SDK's SDKMessage stream into typed HostToWebview messages and
forwarding webview actions back. The webview is a dependency-free TypeScript app that
renders the transcript and composer using VS Code theme variables, so it matches any
color theme.

License

MIT — see LICENSE. The Claude Agent SDK and the Claude runtime it bundles are
licensed separately by Anthropic; your use of them is governed by Anthropic's terms.

Reviews (0)

No results found