feat(commands): unified slash command dispatch + OCR utility commands (#164)

* feat(commands): add OCR utility commands, fix /translate default, defer pending-image submits, fix TipBar clipping

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

* refactor(commands): route all slash commands through unified pre-flight gate

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

* docs(claude): slim slash command flow note to point at docs/commands.md

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

* feat(ux): clarify OCR-supported commands in capability warnings, tips, and docs

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

* fix(ux): scope capability strip link to just the OCR-supported-commands phrase

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

---------

Signed-off-by: Logan Nguyen <lg.131.dev@gmail.com>

Author: quiet-node

CLAUDE.md

@@ -60,6 +60,10 @@ The UI morphs between two states: a compact spotlight-style input bar → an exp
 - **`config/commands.ts`** — slash command registry: defines supported commands and the `SCREEN_CAPTURE_PLACEHOLDER` sentinel used to show a loading tile in chat while a `/screen` capture is in flight
 - **`components/CommandSuggestion.tsx`** — slash command autocomplete popover. Contains `iconForTrigger()`, a switch statement mapping trigger strings to inline SVG constants. **Every new slash command needs a dedicated case here.** Without it, the command falls through to the default, which returns `SCREEN_ICON` (the monitor icon). Steps: (1) add a hoisted `const FOO_ICON = (<svg .../>)` constant, (2) add `case '/foo': return FOO_ICON;` to `iconForTrigger()`.
 
+### Slash commands
+
+User-facing reference for all commands lives in `docs/commands.md`. **Any new slash command must go through the same unified dispatch flow as the existing ones in `src/App.tsx`** (shared pre-flight in `handleSubmit`, then a command-specific stage-2 handler). Do not add a bespoke submit path; extend the existing dispatch instead. This keeps gating, deferral, capability checks, and cancellation behavior consistent across every command.
+
 ### Backend (`src-tauri/src/`)
 
 - **`lib.rs`** — app setup: loads `AppConfig` via `config::load`, converts window to NSPanel (fullscreen overlay), registers tray, spawns hotkey listener, intercepts close events (hides instead of quits)

docs/commands.md

@@ -14,6 +14,10 @@ Commands that operate on text follow a consistent input priority:
 
 This means you can highlight text anywhere on screen, summon Thuki with double-tap Control, type a command, and hit Enter without retyping the selected content.
 
+## Image input on text-only models
+
+`/extract`, `/tldr`, `/translate`, `/rewrite`, `/refine`, `/bullets`, `/todos`, and `/explain` read attached images locally via macOS Vision OCR, so they work even when the active model has no vision capability. Only plain submits and `/screen` alone require a vision model to read images. See [OCR-supported commands](./ocr-commands.md) for the full list and details.
+
 ## /search
 
 Runs agentic web search and answers from live sources with citations.
@@ -97,9 +101,11 @@ Translates text to another language.
 
 **Behavior:** Outputs only the translation with no commentary or explanation.
 
+**Composable:** `/translate` works with attached images or `/screen`. Vision OCR extracts the text first; translation runs on the result. Omitting a target language defaults to Vietnamese.
+
 **Language format:** The target language can be a full name (`French`), ISO code (`fr`, `fra`), or common shorthand.
 
-**Default behavior:** If no language is specified, non-English input translates to English and English input translates to Vietnamese.
+**Default behavior:** If no language is specified, the text is translated to Vietnamese.
 
 ---
 
@@ -115,6 +121,8 @@ Rewrites text to read more naturally and clearly.
 
 **Behavior:** Preserves the original meaning while improving flow and readability. Outputs only the rewritten text.
 
+**Composable:** `/rewrite` works with attached images or `/screen`. Vision OCR extracts the text first, then rewrites it.
+
 ---
 
 ## /tldr
@@ -129,6 +137,8 @@ Summarizes text into 1-3 short, direct sentences.
 
 **Behavior:** Captures the core message, key decision, or critical takeaway. Skips background detail and qualifications.
 
+**Composable:** `/tldr` works with attached images or `/screen`. Vision OCR extracts the text first, then summarizes it.
+
 ---
 
 ## /refine
@@ -143,6 +153,8 @@ Fixes grammar, spelling, and punctuation while preserving your voice.
 
 **Behavior:** Corrects errors and smooths rough phrasing without restructuring or adding new ideas. Your original tone and meaning stay intact.
 
+**Composable:** `/refine` works with attached images or `/screen`. Vision OCR extracts the text first, then refines it.
+
 ---
 
 ## /bullets
@@ -157,6 +169,8 @@ Extracts key points from text as a markdown bullet list.
 
 **Behavior:** Each point is a concise, self-contained statement. Ordered by importance or logical sequence. Filler and repetition are removed. Output uses `- ` prefixed markdown bullets.
 
+**Composable:** `/bullets` works with attached images or `/screen`. Vision OCR extracts the text first, then extracts key points.
+
 ---
 
 ## /explain
@@ -172,6 +186,8 @@ Explains any concept, term, or code snippet in plain language, always with a con
 
 **Behavior:** Outputs a brief explanation followed by at least one concrete example. Assumes no background knowledge. Skips jargon or defines it when unavoidable. No intro or sign-off.
 
+**Composable:** `/explain` works with attached images or `/screen`. Vision OCR extracts the text first, then explains it.
+
 ---
 
 ## /todos
@@ -185,3 +201,5 @@ Summarizes what a piece of text is about, then extracts every task, action item,
 - `/todos [paste a conversation or notes]`: processes typed or pasted content
 
 **Behavior:** Responds in two parts: a short paragraph explaining the context and what is at stake, followed by a `- [ ]` checkbox list of all tasks. Each to-do includes who is responsible, plus any deadline or timeframe if mentioned.
+
+**Composable:** `/todos` works with attached images or `/screen`. Vision OCR extracts the text first, then extracts to-dos.