Add Codex Responses support and session debugging

Teach provider config and auth resolution about the Codex Responses
dialect, including user-facing `style = "openai_responses"` with
`dialect = "codex"`. Serialize and parse Responses traffic with
provider-specific reasoning replay, assistant phase metadata, and robust
function-call assembly keyed by `output_index` so streamed tool inputs
survive proxy quirks and empty terminal payloads.

Also persist thinking origins and message metadata across sessions, add
the Anthropic interleaved-thinking header switch, write per-session
debug logs, and improve the TUI and scripts for inspecting tool output
and session costs.

1 files changed, 68 insertions, 1 deletions

diff --git a/src/config_file.zig b/src/config_file.zig
index 1823177..8f6725c 100644
--- a/src/config_file.zig
+++ b/src/config_file.zig
@@ -77,6 +77,10 @@ pub const Provider = struct {
     /// the name used on the left of a `<provider>:<model>` reference.
     name: []const u8,
     style: APIStyle,
+    /// User-facing sub-dialect for styles that share a protocol family.
+    /// Currently only `style = "openai_responses"` supports `dialect = "codex"`,
+    /// which maps to internal `APIStyle.openai_codex_responses`.
+    dialect: ?[]const u8 = null,
     base_url: []const u8,
     /// The `auth.<name>` session this provider draws its credential from.
     auth_name: []const u8,
@@ -90,6 +94,7 @@ pub const Provider = struct {
 
     pub fn deinit(self: Provider, alloc: Allocator) void {
         alloc.free(self.name);
+        if (self.dialect) |d| alloc.free(d);
         alloc.free(self.base_url);
         alloc.free(self.auth_name);
         // `extra_headers` lives in the Config auth arena; not freed here.
@@ -213,6 +218,14 @@ pub fn buildProviderConfig(
             .max_tokens = max_tokens,
             .extra_headers = prov.extra_headers,
         } },
+        .openai_codex_responses => return .{ .openai_codex_responses = .{
+            .api_key = api_key,
+            .base_url = prov.base_url,
+            .model = wire_model,
+            .reasoning = reasoning,
+            .max_tokens = max_tokens,
+            .extra_headers = prov.extra_headers,
+        } },
     }
 }
 
@@ -677,7 +690,16 @@ fn resolveProvider(
 
     const style_str = (val.get("style") orelse return error.InvalidProvider).asString() orelse
         return error.InvalidProvider;
-    const style = std.meta.stringToEnum(APIStyle, style_str) orelse return error.InvalidProvider;
+    var style = std.meta.stringToEnum(APIStyle, style_str) orelse return error.InvalidProvider;
+    if (style == .openai_codex_responses) return error.InvalidProvider;
+    const dialect = if (tomlStr(val, "dialect")) |d| d else null;
+    if (dialect) |d| {
+        if (std.mem.eql(u8, style_str, "openai_responses") and std.mem.eql(u8, d, "codex")) {
+            style = .openai_codex_responses;
+        } else {
+            return error.InvalidProvider;
+        }
+    }
 
     const base_url = (val.get("base_url") orelse return error.InvalidProvider).asString() orelse
         return error.InvalidProvider;
@@ -699,6 +721,7 @@ fn resolveProvider(
     return .{
         .name = try allocator.dupe(u8, name),
         .style = style,
+        .dialect = if (dialect) |d| try allocator.dupe(u8, d) else null,
         .base_url = try allocator.dupe(u8, base_url),
         .auth_name = try allocator.dupe(u8, auth_name),
         .prompt_cache = prompt_cache,
@@ -1231,6 +1254,50 @@ test "resolve: provider extra_headers flow into the built provider config" {
     try testing.expectEqualStrings("tok", pc.openai_chat.api_key);
 }
 
+test "resolve: openai_responses codex dialect maps to internal API style" {
+    const a = testing.allocator;
+    var env = emptyEnv(a);
+    defer env.deinit();
+    const src =
+        \\[providers.codex]
+        \\style = "openai_responses"
+        \\dialect = "codex"
+        \\base_url = "https://chatgpt.com/backend-api"
+        \\auth = "k"
+        \\
+        \\[auth.k]
+        \\key = "tok"
+    ;
+    const doc = try parseDoc(a, src);
+    defer doc.deinit();
+    var cfg = try resolve(a, &env, doc.root);
+    defer cfg.deinit();
+
+    try testing.expectEqual(APIStyle.openai_codex_responses, cfg.provider("codex").?.style);
+    var defs = models_toml.ModelRegistry.init(a);
+    defer defs.deinit();
+    const pc = try buildProviderConfig(&cfg, &defs, .{ .provider = "codex", .model = "gpt-5.5" });
+    try testing.expectEqual(APIStyle.openai_codex_responses, pc.style());
+}
+
+test "resolve: internal openai_codex_responses style is not user-facing" {
+    const a = testing.allocator;
+    var env = emptyEnv(a);
+    defer env.deinit();
+    const src =
+        \\[providers.codex]
+        \\style = "openai_codex_responses"
+        \\base_url = "https://chatgpt.com/backend-api"
+        \\auth = "k"
+        \\
+        \\[auth.k]
+        \\key = "tok"
+    ;
+    const doc = try parseDoc(a, src);
+    defer doc.deinit();
+    try testing.expectError(error.InvalidProvider, resolve(a, &env, doc.root));
+}
+
 test "resolve: prompt_cache defaults true and is parsed when set" {
     const a = testing.allocator;
     var env = emptyEnv(a);