Build the system prompt from config. Delegates to `claurst_core::system_prompt::build_system_prompt` so that all default content (capabilities, safet

Return all built-in output styles in display order.

Resolve the effective max-tokens.

Only the enabled plugins.

Evaluate whether `tool_name` should be allowed to run. Evaluation order (faithful to TS behaviour): 1. BypassPermissions → always Allow. 2. Check den

Number of models in the registry.

List all sessions, sorted by most-recently-updated first.

Discover and load all plugins from the standard locations. Search order: 1. `~/.claurst/plugins/` (user-global) 2. `<project_dir>/.claurst/plugins/`

Default cache filename — derived from the source URL so a custom `CLAURST_MODELS_URL` doesn't stomp the canonical models.dev cache.

Open (or create) the goal database.

Parse the model's response into `ExtractedMemory` entries. Expected line format: `MEMORY: <category> | <confidence 0-10> | <fact>`

Process a single notification message from the transport stream. Routes resource updates to subscribers and logs other notifications.

Register an LSP server configuration.

Render a system API error block (red-bordered, first 5 lines with [expand] hint, optional retry countdown).

Render a tool result (success variant) — generic fallback.

Resolve the API key for the active provider.

Find the first server name that handles `file_path` based on extension.

Switch to rename mode, pre-populating the input with the current title.

Remove all expired notifications. Call this once per render frame.

Returns true if the token requires Bearer-style authorization (i.e. Claude.ai subscription with `user:inference` scope).

Wrap `text` to fit within `width` display columns, preferring whitespace breaks but falling back to a hard character break when a single token is long

Add a message directly (e.g. from a non-streaming source).

Determine token-warning state given current input token count and model. Thresholds (mirrors TypeScript autoCompact.ts): ≥ 95 % → Critical (red warni

Mark a task as `Cancelled`. No-op if unknown or already terminal.

Compute (current_streak, longest_streak) in days from the aggregated stats. A streak is a consecutive run of calendar days with any activity, ending o

Return the effective context-window size in tokens for the given model. These are approximate; the API enforces the real limits server-side.

Send a streaming `POST /v1/messages`. Events are dispatched to the provided `handler` in real time, and also forwarded into the returned channel so t

Detect if a sequence of lines forms a markdown table

Discover all skills from all configured sources. Returns a `HashMap` of `skill_name → DiscoveredSkill` (first match wins; duplicates from lower-prior

Dismiss the context menu.

Resolve the effective output style for system-prompt assembly.

If the requested id already exists, suffix with -2, -3, … until free.

Expand `${VAR_NAME}` and `${VAR_NAME:-default}` patterns in `input` using the process environment. Unknown variables without a default are left as-is

Build the compaction prompt, optionally with custom instructions appended.

Run a git command in `repo_root` and return stdout as a String. Returns empty string on failure (non-zero exit, not-a-repo, etc.).

Get the cached height for item `idx` at `width`, computing it if needed.

Load user-defined output styles from a directory. Supported file formats: - `.md` — Markdown: `# Label\ndescription\n\nprompt text…` - `.json` — JS

Get models for a provider from the model registry (models.dev data). Builds picker entries from the bundled / network-refreshed registry. The registr

Parse a skill markdown file. Expects optional YAML frontmatter delimited by `---`. Returns `None` when the file is empty after trimming.

Handle a paste event.

Render a collapsed read/search tool use summary. Shows: `▸ ` in yellow + `{tool_name} ` in yellow bold + first few paths comma-joined, followed by `(+

Render the DialogSelect overlay — OpenCode-style: dark overlay, no border, full-width highlight bar on selected item, minimal and polished.

Render a file-edit/write tool result: `Updated file` or `Created file`.

Render a transcript task assignment row using the same structured title/subtitle language.

Resolve the prompt text for the selected output style, including user-defined styles loaded from `~/.claurst/output-styles/`.

Send a JSON-RPC notification (fire-and-forget, no response expected).

Send a JSON-RPC request and wait for the matching response.

Return status for a single server.

Show the dialog (called at startup when bypass mode is active).

Show the notification for the given memory file path. After `dismiss()` is called, re-showing is allowed (unlike the upsell banners which are session-

Return all built-in slash commands.

Return all styles available for `config_dir`: built-ins first, then styles from `<config_dir>/output-styles/`. `config_dir` is typically `~/.claurst`

Return a reference to all registered tips.

Append a line of output to an existing task. No-op if unknown.

Compute typeahead suggestions for slash commands only (e.g., `/help`).

Confirm the current selection; advances to the `Confirming` step. Returns the selected message index if in the Selecting step.

Connect to Chrome at the given port. Picks the first available target (tab/page).

Compute a stable device fingerprint from machine-local information. Combines hostname, login user name, and home directory path, then SHA-256 hashes

Spawn and initialize the server for `file_path` if it is not already running. Returns `None` when no server is configured for this file type.

Expand a skill's prompt template, substituting `$ARGUMENTS` and `$ARGUMENTS_SUFFIX`. - `$ARGUMENTS` → replaced by `args` verbatim (or `""` whe

Backspace in the filter.

Build the explanation paragraph shown in the permission dialog. Mirrors the TS `createPermissionRequestMessage` / `permissionExplainer` output style.