Claudian

To use Claudian, you must meet the following requirements:

1. AI Agent Harness (At least one):

• Claude Code CLI
• Codex CLI
• Grok Build
• OpenCode
• Pi

2. API Provider/Subscription: A compatible subscription or API provider (e.g., OpenRouter, Kimi, GLM, or DeepSeek).

3. Software & Platform:

• Obsidian v1.7.2+
• Desktop only (macOS, Linux, Windows)

To install Claudian from source for development purposes, clone the repository into your vault's plugins folder and build it using npm.

1. Clone the repository:

cd /path/to/vault/.obsidian/plugins
git clone https://github.com/YishenTu/claudian.git
cd claudian

2. Install dependencies and build:

npm install
npm run build

3. Enable the plugin in Obsidian via Settings → Community plugins.

The recommended way to install Claudian is through the Obsidian Community Plugins browser:

1. Open Obsidian → Settings → Community plugins → Browse.
2. Search for "Claudian" and click Install.
3. Enable the plugin.

The Codex provider uses specific trigger characters to distinguish between system commands and user skills in the command dropdown:

• Commands: Triggered using the / prefix (e.g., /compact).
• Skills: Triggered using the $ prefix (e.g., $skill-name).

Built-in commands like compact are available under the / prefix, while skills discovered from the Codex app server are mapped to the $ prefix.

/* Example of how triggers are categorized */
// Commands (Built-in)
/compact

// Skills (User/Repo)
$skill-name

Claude Code (CC) compatible permission rules are defined as strings that specify which tools are allowed, denied, or require confirmation.

Rules follow the format Tool(pattern) or simply Tool to apply to all actions of that tool.

Examples:

• Bash(git *): Allows the Bash tool for any command starting with git .
• Read(*.md): Allows the Read tool for all Markdown files.
• WebFetch(domain:github.com): Allows the WebFetch tool specifically for the github.com domain.
• Read: Allows the Read tool for everything.

// Use createPermissionRule to cast a string to a PermissionRule
const rule = createPermissionRule("Bash(git *)");

When probeRuntimeCommands discovers commands from the Claude SDK, they are mapped to the internal SlashCommand format. Each discovered command includes:

• id: A unique identifier prefixed with sdk: (e.g., sdk:name).
• name: The original command name from the SDK.
• description: The command's description.
• argumentHint: Information about expected arguments.
• content: An empty string (as these are discovered via the SDK probe).
• source: Set to 'sdk' to indicate the origin.

The TabBar supports expanding tab titles to show the full text instead of just the index number. This is toggled via a double-click on a badge.

• Double-click: Toggles the expansion state of a specific tab.
• State Persistence: You can control which tabs are expanded using setExpandedTitleTabIds(tabIds: readonly TabId[]).
• Retrieval: Use getExpandedTitleTabIds() to see which tabs are currently in an expanded state.

When a title is expanded, it is truncated to a maximum length of 32 characters followed by ... if it exceeds that limit.

The ProviderCommandDiscoverySnapshot<T> represents the current state of the command discovery process. It can be one of the following:

• { status: 'idle' }: The store is in its initial state or has been invalidated.
• { status: 'loading' }: A discovery process is currently in progress.
• ProviderCommandDiscoveryResult<T>: The discovery process has completed. If the status is 'ready', it contains the discovered items. If the status is 'error', it contains an error message and a retryable boolean.

The permissionMode setting determines how Claudian handles tool execution requests. It can be set to one of three values:

• 'yolo': Likely allows automatic execution without manual intervention.
• 'plan': Likely requires a plan or confirmation before execution.
• 'normal': The standard permission level.

Claudian allows you to hide specific provider commands via the hiddenProviderCommands configuration object in your ClaudianSettings. This object maps a providerId to an array of command names that should be suppressed from the UI.

Configuration Structure

The hiddenProviderCommands object follows this shape:

{
  "providerId": ["commandName1", "commandName2"]
}

Normalization Rules

When providing command lists, the system applies the following normalization:

• Trimming: Leading and trailing whitespace is removed.
• Prefix Stripping: Leading / or $ characters are stripped from the command name.
• Uniqueness: Duplicate commands (case-insensitive) are removed.
• Empty Values: Providers with no valid commands are omitted from the final configuration.