Download remote settings and memory files. Returns `Ok(None)` when the server has no data for this user (404). Fails open — callers should treat erro

Ensure the selected item is within the scrolling viewport.

Estimate token count from text using simple heuristic.

Rough token count heuristic (chars / 4 + JSON overhead). Mirrors the TS `countTokens` approximation.

Execute the currently selected context menu item.

Execute a single tool invocation.

Expand @include directives in content. Circular references are detected via `visited` set.

Extract key memories from a conversation. Calls the API with a structured extraction prompt and parses the response into `ExtractedMemory` entries.

Extract chatgpt-account-id from the JWT access token. The account_id is in the middle segment (payload) under https://api.openai.com/auth.account_id

Split text that may contain `<file path="...">...</file>` injection blocks into alternating Plain and FileBlock segments.

Pulls the objective text out of the `args` portion of a `/goal …` slash command. Returns `None` for empty args or for the subcommand forms (`status`,

Recognizes a raw `/goal <objective>` user message. Returns the objective string when the first line is `/goal …` with actual objective text; returns `

Extract a list of `"path:line"` strings from an LSP `Location | Location[]` result.

Eye-row style: pink text on black background.

Eyeball highlight style: white on black.

Feed one line (without the trailing newline). Returns `Some(frame)` when a blank line signals the end of an event.

Filter the tool list based on the agent's access level. - "full" → all tools allowed (no filtering) - "read-only" → only ReadOnly/None permis

Find indices of items matching `query` (case-insensitive substring).

Look up a named command by its primary name (case-insensitive).

Fireworks AI — fast inference. Reads `FIREWORKS_API_KEY`.

Format memory headers as a text manifest: one entry per file with `[type] filename (iso-timestamp): description`. Mirrors `formatMemoryManifest` in `

Build the text output for `/plugin list`.

Friendli — serverless inference. Reads `FRIENDLI_TOKEN`.

Generate a cryptographically random code verifier (43–128 chars of Base64url characters, as required by RFC 7636). Uses `getrandom` via the `rand` cr

Get an entry by `"provider_id/model_id"` key.

Get the cache directory for web_fetch content.

Get cached diagnostics for `file_path`.

Get cached diagnostics for `file_path` across all running servers.

Return all recorded modifications for `path`, in chronological order.

Get IDE context from the lockfile (if an IDE is connected). Returns a formatted string like: `IDE: VS Code, workspace: /path/to/project, selection: L

Return the OAuth config appropriate for the current environment. Free-code always uses production OAuth. The `USER_TYPE=ant` gate and staging variant

Return references to all `Thinking` blocks in this message.

Build the text appended to the dynamic section of the system prompt when a goal is active. This is NOT cached (it changes per session).

Returns true when the /goal feature is enabled. Disabled only if CLAURST_GOALS=0 is set explicitly.

Handle a context menu action.

Stop PTT recording: flip the AtomicBool inside VoiceRecorder so the capture thread exits, then fire a "Transcribing…" notice. The transcript text arr

Returns `true` when the provider has no usable API key.

Hugging Face Inference API. Reads `HF_TOKEN`.

Perform the LSP `initialize` / `initialized` handshake.

Handle slash commands that should open UI screens rather than execute as normal commands. Returns `true` if the command was intercepted.

Whether the auto-memory system is enabled for this session. Priority chain (mirrors `isAutoMemoryEnabled` in `paths.ts`): 1. `CLAURST_DISABLE_AUTO_ME

Whether `provider_id` refers to the OpenAI Codex provider under either of its two id spellings: the canonical `"codex"` (used by `CodexProvider`, the

Returns `true` if `message` matches any known context-overflow pattern. The comparison is case-insensitive. Patterns are matched as substrings (not

Whether a plugin is enabled.

Return `true` when `val` is a truthy env-var value. Truthy: `"1"`, `"true"`, `"yes"`, `"on"` (case-insensitive). `None` (variable unset) is falsy.

Returns `true` if the token is expired or will expire within `grace_secs`.

Whether `id` is a GPT-5 reasoning model (`gpt-5*`), excluding the non-reasoning chat / pro snapshots that ignore `reasoning_effort`.

Compare two semver strings. Returns `true` when `latest` > `current`.

Whether this state represents an active/visible connection that occupies horizontal space in the status bar (i.e. not Disconnected).

Return `true` if this banner should be shown.

Returns a JSON-compatible value for this field's current state.

Scroll to the very bottom.

Ctrl+U: Cut from line start to cursor and save to kill ring.

Ctrl+W: Cut previous word and save to kill ring.

List every entry in the registry.

List all installed plugins.

Load cached flags from disk

Load a companion soul from `{config_dir}/companion.json`. Returns `None` if the file does not exist or cannot be parsed.

Load a single AGENTS.md file (respects MAX_FILE_SIZE, expands @includes).

Load and truncate the `MEMORY.md` index from `memory_dir`. Returns `None` when the file does not exist or is empty. Mirrors the entrypoint-reading pa

Parse a skill Markdown file into a `SkillDefinition`. Expected format: ```text --- name: my-skill description: Does something useful tags: [tag1, tag

Load persisted state; returns `Default` on any error (missing file, parse failure).

Convert text region to lowercase.

Map an HTTP error response to a typed ProviderError.

Typed variant of `match_session_mode` — accepts an `AgentMode` directly. Returns a warning string when the environment was changed, `None` otherwise.

If the active permission dialog's selected option is the prefix-allow option ('P') for a Bash dialog, extract the suggested prefix and add it to `bash