Plug the deep-embedder holes; remove public.zig escape hatch

Close the gap that forced the CLI to import internal libpanto namespaces,
then delete the transitional re-exports. The CLI now uses only the curated
public surface.

Additions to the public Agent facade:
- init/deinit (heap-pin the inner; the handle is a copyable value).
- addSystemMessage (.append) and setSystemPrompt (.replace).
- compact(override_system_prompt, extra) falling back to the config prompt.
- sessionId() accessor.

CompactionConfig gains compaction_prompt, which owns the compaction system
prompt for both auto-compaction and the explicit compact() default; the
Agent.compaction_system_prompt field is deleted. ConversationData is the
public name for the owned conversation value type (Session.load returns it,
Agent.init adopts it), distinct from the borrowed Conversation handle.

Agent.addSystemMessage/setSystemPrompt are kept on Agent (they persist with
the SystemMode) rather than dropped as the plan first proposed.

4 files changed, 73 insertions, 64 deletions

diff --git a/src/command.zig b/src/command.zig
index 49ad5b8..0f6fb01 100644
--- a/src/command.zig
+++ b/src/command.zig
@@ -32,10 +32,10 @@ const panto = @import("panto");
 pub const Context = struct {
     allocator: std.mem.Allocator,
 
-    /// The active agent. Owns the conversation (`agent.conversation`) and
-    /// the session store (`agent.session_store`); commands reach both
-    /// through it rather than holding separate pointers.
-    agent: *panto.agent.Agent,
+    /// The active agent (a copyable public handle). Commands reach the
+    /// conversation via `agent.conversation()` and compact via
+    /// `agent.compact(...)`.
+    agent: panto.Agent,
 
     /// REPL output writer and its backing file writer (for flushing).
     stdout: *std.Io.Writer,
diff --git a/src/compaction.zig b/src/compaction.zig
index 4ec5732..5d64f98 100644
--- a/src/compaction.zig
+++ b/src/compaction.zig
@@ -22,7 +22,7 @@ pub fn register(registry: *command.Registry) !void {
 fn run(args: []const u8, ctx: *command.Context) anyerror!void {
     const extra: ?[]const u8 = if (args.len > 0) args else null;
 
-    const res = ctx.agent.compactAndPersist(ctx.compaction_prompt, extra) catch |err| {
+    const res = ctx.agent.compact(ctx.compaction_prompt, extra) catch |err| {
         try ctx.stdout.print("\n[compaction failed: {s}]\n> ", .{@errorName(err)});
         try ctx.stdout_file.flush();
         return;
diff --git a/src/main.zig b/src/main.zig
index 995735f..61a88bd 100644
--- a/src/main.zig
+++ b/src/main.zig
@@ -144,7 +144,7 @@ const CLIRenderer = struct {
 /// it terminates (`turn_complete` → `next()` returns null). A failure
 /// surfaces as the error from `next()`; the caller renders it. The stream is
 /// always `deinit`ed (persisting the turn tail) on every exit path.
-fn driveTurn(agent: *panto.agent.Agent, message: panto.agent.Agent.UserMessage, renderer: *CLIRenderer) !void {
+fn driveTurn(agent: panto.Agent, message: panto.UserMessage, renderer: *CLIRenderer) !void {
     var stream = try agent.run(message);
     defer stream.deinit();
     while (try stream.next()) |ev| try renderer.render(ev);
@@ -358,7 +358,7 @@ pub fn main(init: std.process.Init) !void {
     // On resume, reconstruct the conversation from the store. (The
     // dangling-prompt recovery feature was dropped in R2.) On a fresh
     // session, the agent starts with an empty conversation.
-    var adopted_conversation: ?panto.conversation.Conversation = null;
+    var adopted_conversation: ?panto.ConversationData = null;
     if (is_resume) {
         adopted_conversation = try session.load();
         const sid = session.info.id;
@@ -376,11 +376,11 @@ pub fn main(init: std.process.Init) !void {
     // built but before the first turn, so in-place registration is visible.
     // System-prompt seeding/reconciliation below runs *through* the agent
     // so those entries persist.
-    const active_config: panto.Config = .{
+    var active_config: panto.Config = .{
         .provider = provider_config,
         .compaction = compaction_cfg,
     };
-    var agent = panto.agent.Agent.init(
+    const agent = try panto.Agent.init(
         alloc,
         io,
         &active_config,
@@ -424,7 +424,7 @@ pub fn main(init: std.process.Init) !void {
             io,
             init.environ_map,
             luarocks_rt.layout.agent_dir,
-            &agent,
+            agent,
         );
     } else {
         // Fresh session — source and install the system prompt.
@@ -433,7 +433,7 @@ pub fn main(init: std.process.Init) !void {
             io,
             init.environ_map,
             luarocks_rt.layout.agent_dir,
-            &agent,
+            agent,
         );
     }
 
@@ -476,7 +476,9 @@ pub fn main(init: std.process.Init) !void {
         init.environ_map,
         luarocks_rt.layout.agent_dir,
     );
-    agent.compaction_system_prompt = compaction_prompt;
+    // The compaction prompt is resolved after extension load; set it on the
+    // config the agent re-reads each turn (visible before the first turn).
+    active_config.compaction.compaction_prompt = compaction_prompt;
 
     const banner_base: []const u8 = switch (provider_config) {
         inline else => |c| c.base_url,
@@ -521,7 +523,7 @@ pub fn main(init: std.process.Init) !void {
 
     var cmd_ctx: command.Context = .{
         .allocator = alloc,
-        .agent = &agent,
+        .agent = agent,
         .stdout = stdout,
         .stdout_file = &stdout_file,
         .compaction_prompt = compaction_prompt,
@@ -570,7 +572,7 @@ pub fn main(init: std.process.Init) !void {
         // auto-compaction) through its session store — the CLI no longer
         // touches persistence.
         cli_renderer.beginTurn();
-        driveTurn(&agent, .{ .text = line }, &cli_renderer) catch |err| {
+        driveTurn(agent, .{ .text = line }, &cli_renderer) catch |err| {
             cli_renderer.renderError(err);
             try stdout.print("\n[error: {s}]\n", .{@errorName(err)});
         };
diff --git a/src/system_prompt.zig b/src/system_prompt.zig
index 500b135..39c3ea8 100644
--- a/src/system_prompt.zig
+++ b/src/system_prompt.zig
@@ -137,7 +137,7 @@ pub fn seedFresh(
     io: Io,
     environ_map: *const std.process.Environ.Map,
     base_dir: []const u8,
-    agent: *panto.agent.Agent,
+    agent: panto.Agent,
 ) !void {
     const user_dir = try userLayerDir(arena, environ_map);
     const project_dir = try projectLayerDir(arena, io);
@@ -151,13 +151,13 @@ pub fn seedFreshLayers(
     base_dir: ?[]const u8,
     user_dir: ?[]const u8,
     project_dir: ?[]const u8,
-    agent: *panto.agent.Agent,
+    agent: panto.Agent,
 ) !void {
     const resolved = try resolveLayers(arena, io, base_dir, user_dir, project_dir);
     const blocks = try resolved.blocks(arena);
     for (blocks) |text| {
         // The agent adds to its conversation and persists the entry.
-        try agent.addSystemMessage(text, .append);
+        try agent.addSystemMessage(text);
     }
 }
 
@@ -174,7 +174,7 @@ pub fn reconcileResume(
     io: Io,
     environ_map: *const std.process.Environ.Map,
     base_dir: []const u8,
-    agent: *panto.agent.Agent,
+    agent: panto.Agent,
 ) !void {
     const user_dir = try userLayerDir(arena, environ_map);
     const project_dir = try projectLayerDir(arena, io);
@@ -188,21 +188,21 @@ pub fn reconcileResumeLayers(
     base_dir: ?[]const u8,
     user_dir: ?[]const u8,
     project_dir: ?[]const u8,
-    agent: *panto.agent.Agent,
+    agent: panto.Agent,
 ) !void {
     const resolved = try resolveLayers(arena, io, base_dir, user_dir, project_dir);
     const config_blocks = try resolved.blocks(arena);
 
-    const window = try effectiveConfigWindow(arena, agent.conversation.messages.items);
+    const window = try effectiveConfigWindow(arena, agent.conversation().messages());
 
     if (blocksEqual(config_blocks, window)) return;
 
-    // Re-seed: one `replace` for the seed, then one `append` per append.
-    // The agent adds to its conversation and persists each entry (with the
-    // correct system mode).
-    try agent.addSystemMessage(config_blocks[0], .replace);
+    // Re-seed: one `setSystemPrompt` (replace) for the seed, then one
+    // `addSystemMessage` (append) per append. The agent adds to its
+    // conversation and persists each entry (with the correct system mode).
+    try agent.setSystemPrompt(config_blocks[0]);
     for (config_blocks[1..]) |text| {
-        try agent.addSystemMessage(text, .append);
+        try agent.addSystemMessage(text);
     }
 }
 
@@ -339,7 +339,7 @@ test "blocksEqual - length and content" {
 
 test "effectiveConfigWindow - no replace anchors to leading system blocks" {
     const alloc = testing.allocator;
-    var conv = panto.conversation.Conversation.init(alloc);
+    var conv = panto.ConversationData.init(alloc);
     defer conv.deinit();
 
     try conv.addSystemMessage("seed");
@@ -357,7 +357,7 @@ test "effectiveConfigWindow - no replace anchors to leading system blocks" {
 
 test "effectiveConfigWindow - anchors to last replace block" {
     const alloc = testing.allocator;
-    var conv = panto.conversation.Conversation.init(alloc);
+    var conv = panto.ConversationData.init(alloc);
     defer conv.deinit();
 
     try conv.addSystemMessage("old seed");
@@ -473,38 +473,45 @@ fn openTmpStore(arena: Allocator, root: []const u8) !panto.FileSystemJSONLStore
 /// harness only holds the config to keep the agent's borrowed pointer valid.
 const SPAgentHarness = struct {
     config: panto.Config,
-    agent: panto.agent.Agent,
+    agent: panto.Agent,
+    /// A second handle onto the same session (a copyable value proxying to
+    /// the same store + id) so the test can `forceFlush` without reaching
+    /// agent internals. `session_id` mirrors the agent's session id.
+    session: panto.Session,
+    session_id: []const u8,
 
     fn init(
         self: *SPAgentHarness,
         session: panto.Session,
-        adopted: ?panto.conversation.Conversation,
-    ) void {
+        adopted: ?panto.ConversationData,
+    ) !void {
         self.config = .{
             .provider = .{ .openai_chat = .{ .api_key = "k", .base_url = "u", .model = "m" } },
         };
-        self.agent = panto.agent.Agent.init(testing.allocator, testing.io, &self.config, session, adopted);
+        self.session = session;
+        self.session_id = session.info.id;
+        self.agent = try panto.Agent.init(testing.allocator, testing.io, &self.config, session, adopted);
     }
 
     fn deinit(self: *SPAgentHarness) void {
         self.agent.deinit();
     }
-};
 
-/// Force the session to flush to disk by appending an assistant message
-/// through the agent's session. (System/user messages buffer in memory
-/// until the first assistant message; a realistic resume only sees what a
-/// completed turn persisted.)
-fn forceFlush(agent: *panto.agent.Agent) !void {
-    var conv = panto.conversation.Conversation.init(testing.allocator);
-    defer conv.deinit();
-    try conv.addAssistantMessage(&.{});
-    const id: panto.WireIdentity = .{ .api_style = .openai_chat, .base_url = "u", .model = "m" };
-    var batch = [_]panto.PersistentMessage{
-        .{ .message = conv.messages.items[0], .identity = id },
-    };
-    try agent.session.append(&batch);
-}
+    /// Force the session to flush to disk by appending an assistant message
+    /// through the (shared) session handle. (System/user messages buffer in
+    /// memory until the first assistant message; a realistic resume only
+    /// sees what a completed turn persisted.)
+    fn forceFlush(self: *SPAgentHarness) !void {
+        var conv = panto.ConversationData.init(testing.allocator);
+        defer conv.deinit();
+        try conv.addAssistantMessage(&.{});
+        const id: panto.WireIdentity = .{ .api_style = .openai_chat, .base_url = "u", .model = "m" };
+        var batch = [_]panto.PersistentMessage{
+            .{ .message = conv.messages.items[0], .identity = id },
+        };
+        try self.session.append(&batch);
+    }
+};
 
 test "seedFresh then no-config-change resume is a no-op" {
     const alloc = testing.allocator;
@@ -526,12 +533,12 @@ test "seedFresh then no-config-change resume is a no-op" {
     var session_id_len: usize = 0;
     {
         var h: SPAgentHarness = undefined;
-        h.init(store.create(), null);
+        try h.init(store.create(), null);
         defer h.deinit();
-        try seedFreshLayers(arena, testing.io, null, null, project_dir, &h.agent);
-        try forceFlush(&h.agent); // persist seed (+ assistant) to disk
-        @memcpy(session_id_buf[0..h.agent.session.info.id.len], h.agent.session.info.id);
-        session_id_len = h.agent.session.info.id.len;
+        try seedFreshLayers(arena, testing.io, null, null, project_dir, h.agent);
+        try h.forceFlush(); // persist seed (+ assistant) to disk
+        @memcpy(session_id_buf[0..h.session_id.len], h.session_id);
+        session_id_len = h.session_id.len;
     }
     const session_id = session_id_buf[0..session_id_len];
 
@@ -547,10 +554,10 @@ test "seedFresh then no-config-change resume is a no-op" {
     var sess2 = (try store.resolve(session_id)).?;
     const conv2 = try sess2.load();
     var h2: SPAgentHarness = undefined;
-    h2.init(sess2, conv2);
+    try h2.init(sess2, conv2);
     defer h2.deinit();
-    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, &h2.agent);
-    const eff_after = try panto.effectiveSystemBlocks(arena, h2.agent.conversation.messages.items);
+    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, h2.agent);
+    const eff_after = try panto.effectiveSystemBlocks(arena, h2.agent.conversation().messages());
     try testing.expectEqual(seeded_count, eff_after.items.len);
 }
 
@@ -572,12 +579,12 @@ test "resume after config change appends replace + append sequence" {
     var session_id_len: usize = 0;
     {
         var h: SPAgentHarness = undefined;
-        h.init(store.create(), null);
+        try h.init(store.create(), null);
         defer h.deinit();
-        try seedFreshLayers(arena, testing.io, null, null, project_dir, &h.agent);
-        try forceFlush(&h.agent); // persist old seed (+ assistant) to disk
-        @memcpy(session_id_buf[0..h.agent.session.info.id.len], h.agent.session.info.id);
-        session_id_len = h.agent.session.info.id.len;
+        try seedFreshLayers(arena, testing.io, null, null, project_dir, h.agent);
+        try h.forceFlush(); // persist old seed (+ assistant) to disk
+        @memcpy(session_id_buf[0..h.session_id.len], h.session_id);
+        session_id_len = h.session_id.len;
     }
     const session_id = session_id_buf[0..session_id_len];
 
@@ -588,20 +595,20 @@ test "resume after config change appends replace + append sequence" {
     var sess2 = (try store.resolve(session_id)).?;
     const conv2 = try sess2.load();
     var h2: SPAgentHarness = undefined;
-    h2.init(sess2, conv2);
+    try h2.init(sess2, conv2);
     defer h2.deinit();
-    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, &h2.agent);
+    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, h2.agent);
 
     // The effective prompt now reflects only the new config blocks.
-    const eff = try panto.effectiveSystemBlocks(arena, h2.agent.conversation.messages.items);
+    const eff = try panto.effectiveSystemBlocks(arena, h2.agent.conversation().messages());
     try testing.expectEqual(@as(usize, 2), eff.items.len);
     try testing.expectEqualStrings("new seed", eff.items[0]);
     try testing.expectEqualStrings("new append", eff.items[1]);
 
     // A second no-op resume must not change the effective prompt (anchors
     // to the new `replace` window, not the stale original seed).
-    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, &h2.agent);
-    const eff2 = try panto.effectiveSystemBlocks(arena, h2.agent.conversation.messages.items);
+    try reconcileResumeLayers(arena, testing.io, null, null, project_dir, h2.agent);
+    const eff2 = try panto.effectiveSystemBlocks(arena, h2.agent.conversation().messages());
     try testing.expectEqual(@as(usize, 2), eff2.items.len);
 }