refactor: extract op-ai crate

Relocate the AI chat layer (chat_provider.rs, chat_models.rs,
agent_settings_state.rs) out of openpencil-shell-core into a dedicated
op-ai crate. These three modules are dependency-free transport-free
data shapes — the ChatProvider trait, the ModelEntry catalog type, and
the Cmd+, settings-modal state — so they form a clean wasm32-clean
leaf crate. A fresh lib.rs declares all three as modules; chat_models'
`crate::agent_settings_state::` path stays valid. openpencil-desktop's
chat_*.rs transports + model_discovery.rs now import `op_ai::*`. Pure
relocation, no behaviour change.

crates/op-ai/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "op-ai"
+version.workspace = true
+edition.workspace = true
+rust-version.workspace = true
+license.workspace = true
+description = "OpenPencil AI chat layer — transport-free ChatProvider trait + model catalog + agent-settings state"
+
+[lib]
+name = "op_ai"
+path = "src/lib.rs"
+
+[dependencies]

crates/op-ai/src/lib.rs

@@ -0,0 +1,20 @@
+//! OpenPencil AI chat layer.
+//!
+//! This crate carries the transport-free data shapes for the editor's
+//! chat / agent integration — extracted out of `openpencil-shell-core`
+//! in the Phase 7 strangler reorg:
+//!
+//! - [`chat_provider`] — the `ChatProvider` trait + provider-category
+//!   types (`CliName`, `Provider`, …) and the `EchoProvider` test
+//!   double. Real transports (tokio / reqwest / process-spawn) live
+//!   desktop-side in `openpencil-desktop`.
+//! - [`chat_models`] — the `ModelEntry` model-catalog type.
+//! - [`agent_settings_state`] — state types for the Cmd+, settings
+//!   modal (`AgentSettingsTab`, `AgentProvider`, …).
+//!
+//! The crate is dependency-free and wasm32-clean so both the native
+//! and web shells can build against it.
+
+pub mod agent_settings_state;
+pub mod chat_models;
+pub mod chat_provider;

crates/openpencil-desktop/Cargo.toml

@@ -25,6 +25,11 @@ op-editor-core = { path = "../op-editor-core" }
 # (`mcp.rs` + `mcp/*`) was extracted out of openpencil-shell-core into the
 # op-mcp crate; `mcp_serve.rs` registers tools through `op_mcp::*`.
 op-mcp = { path = "../op-mcp" }
+# Phase 7 strangler reorg: the transport-free AI chat data shapes
+# (`ChatProvider` trait + model catalog + agent-settings state) were
+# extracted into op-ai; the `src/chat_*.rs` real transports + the
+# model-discovery path import them through `op_ai::*`.
+op-ai = { path = "../op-ai" }
 # Canonical `.op` (PenDocument) → shell `Document` loader / adapter.
 # Extracted out of this binary into a shared library crate so library
 # crates (openpencil-shell-native) can reuse the conversion;

crates/openpencil-desktop/src/chat_claude.rs

@@ -21,7 +21,7 @@ use anthropic_agent_sdk::{
     types::{ContentBlock, Message},
     ClaudeAgentOptions, StreamExt,
 };
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, StopReason,
 };
 use tokio::sync::mpsc;

crates/openpencil-desktop/src/chat_copilot.rs

@@ -27,7 +27,7 @@ use github_copilot_sdk::handler::{
 };
 use github_copilot_sdk::types::{MessageOptions, SessionConfig, SessionEvent};
 use github_copilot_sdk::{Client, ClientOptions};
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, StopReason,
 };
 use tokio::sync::mpsc;

crates/openpencil-desktop/src/chat_http_server.rs

@@ -39,7 +39,7 @@
 use std::sync::Arc;
 use std::time::Duration;
 
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, CliName, StopReason,
 };
 use serde::Serialize;

crates/openpencil-desktop/src/chat_runtime.rs

@@ -1,5 +1,5 @@
 //! Real `ChatProvider` impls — the desktop-side companion to the
-//! abstraction in `openpencil_shell_core::chat_provider`. Shell-core
+//! abstraction in `op_ai::chat_provider`. Shell-core
 //! intentionally stays wasm32-clean (no tokio / reqwest / process), so
 //! transports live here on the native binary.
 //!
@@ -27,7 +27,7 @@ use agent::provider::Provider;
 use agent::query::QueryEngine;
 use agent::stream::Event;
 use futures::StreamExt;
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, StopReason,
 };
 use tokio::runtime::{Builder, Runtime};

crates/openpencil-desktop/src/chat_session.rs

@@ -10,7 +10,7 @@
 use std::sync::mpsc::{self, Receiver, TryRecvError};
 use std::thread;
 
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, CliName,
 };
 use openpencil_shell_native::WidgetHostNative;
@@ -207,7 +207,7 @@ pub fn pump(
 #[cfg(test)]
 mod tests {
     use super::*;
-    use openpencil_shell_core::chat_provider::{EchoProvider, StopReason};
+    use op_ai::chat_provider::{EchoProvider, StopReason};
 
     #[test]
     fn session_streams_echo_provider_deltas_to_completion() {

crates/openpencil-desktop/src/chat_subprocess.rs

@@ -33,7 +33,7 @@
 use std::path::PathBuf;
 use std::sync::Arc;
 
-use openpencil_shell_core::chat_provider::{
+use op_ai::chat_provider::{
     ChatDelta, ChatProvider, ChatRequest, CliName, StopReason,
 };
 use tokio::io::{AsyncBufReadExt, AsyncWriteExt, BufReader};

crates/openpencil-desktop/src/model_discovery.rs

@@ -23,8 +23,8 @@ use std::process::{Command, Stdio};
 use std::sync::mpsc::{self, Receiver, TryRecvError};
 use std::time::{Duration, Instant};
 
-use openpencil_shell_core::agent_settings_state::AgentProvider;
-use openpencil_shell_core::chat_models::ModelEntry;
+use op_ai::agent_settings_state::AgentProvider;
+use op_ai::chat_models::ModelEntry;
 
 /// Background model-discovery probe. [`discover_models`] reads a
 /// cache file and spawns a subprocess (`opencode models`, ~1 s),
@@ -79,7 +79,7 @@ impl ModelProbe {
 
 /// Translate a shell-core `ModelEntry` into op-editor-core's.
 fn model_entry_to_ec(m: ModelEntry) -> op_editor_core::ModelEntry {
-    use openpencil_shell_core::agent_settings_state::AgentProvider as ScP;
+    use op_ai::agent_settings_state::AgentProvider as ScP;
     let provider = match m.provider {
         ScP::ClaudeCode => op_editor_core::AgentProvider::ClaudeCode,
         ScP::CodexCli => op_editor_core::AgentProvider::CodexCli,

crates/openpencil-shell-core/src/lib.rs

@@ -16,9 +16,6 @@
 //!   differentiation lives at the canvas viewport / chrome layer
 //!   (single-page + infinite canvas recommended, multi-page also supported).
 
-pub mod agent_settings_state;
-pub mod chat_models;
-pub mod chat_provider;
 // Phase 3 strangler reorg: i18n extracted into the op-i18n crate. Re-exported
 // as `i18n` so `crate::i18n::translate` / `crate::i18n::Locale` paths still resolve.
 pub use op_i18n as i18n;