Nexora Code

An agentic coding assistant with a skill library that measurably improves.

Chat with your codebase, review a selection, fix what the Problems panel is complaining about, write a commit message — and manage the Agent Skills that every coding agent on your machine reads from.

Runs in VS Code, Cursor, Kiro, Antigravity, Windsurf, VSCodium and anything else built on the VS Code extension API. One build, no proposed APIs, no editor-specific forks.

What it does

Chat that knows what you are looking at. The panel attaches your current selection and the active file's errors and warnings automatically, and shows you exactly what it attached before you send. Replies stream, code blocks come with Copy and Insert, and Esc stops a turn.

Editor commands. Right-click a selection for Explain, Review, and Generate Tests. Nexora: Fix Problems in This File sends the diagnostics with the code. Nexora: Write a Commit Message reads your staged diff.

Slash commands. Type / in the composer for the same verbs without leaving the keyboard — /explain, /review, /tests, /fix, /commit, /effort, /new. /skill <name> [task] runs the next turn under one of your installed skills: the picker matches on descriptions as well as names, and the editor reads that SKILL.md off disk and sends it with the turn, because the runtime answering you cannot see your filesystem.

A Skills view. Every skill installed on your machine, grouped by the root it lives in — because "installed but the agent ignores it" is nearly always "installed into a root that agent does not read". Broken symlinks are shown rather than hidden. Scaffold a new skill into the workspace with one command.

Skills that get better. The runtime records whether each skill helped, drafts a revision when one accumulates failures, and promotes the revision only if it beats the incumbent on a regression suite it did not write. Nothing is promoted on a model's own say-so.

Setup

Install the extension.
Run Nexora: Sign In from the command palette and paste your API key. It goes into the editor's keychain, not a settings file.
If you point at a self-hosted runtime, set nexora.baseUrl.

Already used the CLI? Then there is nothing to do — the extension reads the URL and key from ~/.nexora/config.json and will not ask you to log in twice.

Where skills live

Two roots, by convention:

Root	Read by
`.agents/skills`	Cursor, Windsurf, Kiro, Antigravity, Codex, Gemini, OpenCode, Amp, Cline, Aider, Zed, Copilot
`.nexora/skills`	Nexora Code, Claude Code

Install Skill Pack runs the nexora CLI, which writes real files to the shared root and links the other one to it — so editing a skill edits it for every agent at once. The extension offers to install the CLI if it is missing.

Commands

Command
`Nexora: Open Chat`	`Cmd/Ctrl+Shift+U`
`Nexora: Explain Selection`	`Cmd/Ctrl+Alt+E`
`Nexora: Review Selection`
`Nexora: Generate Tests for Selection`
`Nexora: Fix Problems in This File`
`Nexora: Write a Commit Message`	Also in the Source Control title bar
`Nexora: Ask About This File`
`Nexora: Install Skill Pack`
`Nexora: New Skill in This Workspace`
`Nexora: Set Effort Level`	Also the status bar item
`Nexora: Run Diagnostics`	Start here when something is wrong

In the composer, / opens the same list plus /skill. Arrow keys move, Enter or Tab takes the highlighted row, Escape dismisses the popup without losing what you typed.

Settings

Setting	Default
`nexora.baseUrl`	(from the CLI config)	Runtime URL
`nexora.effort`	`standard`	`fast`, `standard`, `high`
`nexora.mode`	`code`	Tool set and system prompt
`nexora.showReasoning`	`false`	Stream the model's reasoning
`nexora.attachSelection`	`true`	Send the selection with a message
`nexora.attachDiagnostics`	`true`	Send errors and warnings
`nexora.contextLineLimit`	`400`	Most lines of context per message
`nexora.skills.scope`	`user`	Where Install Skill Pack writes
`nexora.skills.showUser`	`true`	Show home-directory skills in the tree

Privacy

Your prompt, the attached selection, and the attached diagnostics are sent to the runtime at nexora.baseUrl — which is your own deployment if you self-host. Nothing else leaves the machine. The extension makes no requests to any other host. API keys are stored in the OS keychain via the editor's SecretStorage.

Text in files is treated as data, never as instructions: attachments are fenced and labelled as context so a comment reading "ignore previous instructions" is just a comment.

Something is wrong

Run Nexora: Run Diagnostics. It prints the editor, the runtime URL, which source the API key came from, whether the runtime answers, how many skills the agent can see, and every local skill root.

Full documentation: https://nexora-intelligence.vercel.app/docs