Settings

Settings file not loaded
Provider settings apply to this server session when saved. Workspace and skills path changes may still need a rebuild or server restart.
Workspace Profiles
Save the current workspace and skills paths for quick reuse.
Save common workspace path pairs here, then apply them without opening the raw env editor.
Core Configuration
Used when LLM_PROVIDER is anthropic_api.
Type or paste a folder path, then use the buttons to validate or browse.
Absolute path to your Claude Code CLI workspace directory
Type or paste a folder path, then use the buttons to validate or browse.
Path to skills folder - browse relative to WORKSPACE_ROOT, or type a relative path
Displayed in the browser tab and app header
Claude CLI
Use anthropic_api for deployable API-key mode or claude_code_cli for local subscription mode.
Must be true before the app will call the local Claude CLI.
Use auto to prefer the official native install path, then fall back to PATH. You can also set an executable path or its install folder.
Use auto for Claude Code's built-in auth login. Set claude-login or an explicit helper path only when you intentionally use a custom login helper.
Default profileNot checked
~\.claude
Select a discovered local Claude profile, or use a manual portable path such as ~/.claude-profiles/<profile>.
Additional Variables
No additional variables. Click + Add to define custom ones.
Setup Center
Readiness, project inventory, and local auth checks
V1 Release Readiness
Release readiness unavailable

Data Boundary

What stays local, what leaves only after an explicit action, and what remains manual.

Workspace files
Read locally from configured workspace and skills paths.
The app avoids broad device scans. Skill reads and writes stay limited to the configured workspace and skills directory.
RAG chat
Provider context is sent only when you send a chat message.
The selected provider receives your prompt plus retrieved skill excerpts and citation metadata. Settings does not run generation on page load.
Diagnostics
Exported only by action and scrubbed before packaging.
Diagnostics omit API keys, account identifiers, OAuth paths, raw profile names, home paths, hook commands, and raw provider output.
Manual checks
OS dialogs, login, and account-backed chat stay user-owned.
Manual QA stores only status and timestamp in this browser, or in memory when browser storage is unavailable.

Use Release Readiness for automated blockers. Mark Manual QA Evidence only after the local device or account check was actually performed.

Manual QA Evidence

Track local-only checks that need a visible OS dialog, account auth flow, or human-owned chat action.

0/3
CLI checklistnpm run qa:manual
Not complete3 pending0 need fix0 skipped

Skipped means intentionally not verified; it is not a pass.

Native folder picker
Pending
Open Settings, click Choose folder, then cancel and select a harmless test folder.Visible OS picker opens, cancel is non-destructive, and selected folders update only the intended field.Manual because: The OS folder picker is a native device dialog and must be visually confirmed by the local user.Not recorded
Claude Open Login
Pending
Select the intended Claude profile, click Open Login, then close or cancel the visible auth flow.Login never runs silently, the launcher is visible, and Settings stays sanitized.Manual because: The login action can open a private account-owned Claude auth flow, so automation must not click it.Not recorded
Account-backed chat
Pending
Use an allowed API key or Claude profile, send the release-readiness prompt, and inspect the cited answer.Chat sends only after user action, provider failures are actionable, and no account/auth details leak.Manual because: A real chat sends a user-owned prompt through configured credentials and needs account approval.Not recorded
Stores only status and timestamp in this browser. It never stores account names, profile paths, prompts, screenshots, or auth output.
3 local manual checks are available.
Claude Project Inventory
Claude project inventory unavailable
Setup Doctor unavailable
RAG Index
Loading...
Skill Quality
Quality report unavailable

First Run Checklist

0 of 7 complete, 6 need action
0/7
Next stepWorkspace path valid: Save
1
Workspace path valid
Needs action
Set WORKSPACE_ROOT to an accessible folder.
2
Skills directory valid
Needs action
Set SKILLS_DIR to the folder containing skill markdown files.
3
Index rebuilt
Needs action
Rebuild the index after workspace and skill paths are valid.
4
Provider selected
Needs action
Choose and save an LLM provider.
5
Claude/API auth tested
Needs action
Add a valid Anthropic API key in Settings.
6
First chat readiness
Needs action
Open Chat after provider, auth, and index readiness are resolved.
7
Diagnostics export
Optional
Export a diagnostics bundle when you are ready to archive setup state.
Complete these in order. The checklist reuses Setup Doctor, runtime provider status, index status, CLI smoke-test state, chat readiness, and diagnostics export.

Claude CLI

Local install, profile auth, and smoke-test state
Loading...

Active Values

Sanitized display of saved provider and workspace inputs
ANTHROPIC_API_KEYAnthropic API Key
Missing
(not set)
WORKSPACE_ROOTWorkspace Root Path
Not checked
(not set)
SKILLS_DIRSkills Directory
Not checked
(not set)
NEXT_PUBLIC_APP_TITLEApp Title
Missing
(not set)
LLM_PROVIDERLLM Provider
Missing
(not set)
ENABLE_LOCAL_CLAUDE_CLIEnable Local Claude CLI
Missing
(not set)
CLAUDE_CLI_PATHClaude CLI Command
Missing
(not set)
CLAUDE_LOGIN_COMMANDClaude Login Helper
Missing
(not set)
CLAUDE_CONFIG_DIRClaude Profile
Not checked
(not set)
Config file
Settings file not loaded