tool call resiliency

3 files changed, 53 insertions, 25 deletions

diff --git a/src/config_file.zig b/src/config_file.zig
index 533eb48..f7589fe 100644
--- a/src/config_file.zig
+++ b/src/config_file.zig
@@ -130,6 +130,7 @@ pub fn buildProviderConfig(
     const def_opt = defs.get(ref.provider, ref.model);
     const wire_model: []const u8 = if (def_opt) |d| d.model else ref.model;
     const reasoning: panto.config.ReasoningEffort = if (def_opt) |d| d.reasoning else .default;
+    const max_tokens: u32 = if (def_opt) |d| (d.max_tokens orelse 64_000) else 64_000;
 
     switch (prov.style) {
         .openai_chat => return .{ .openai_chat = .{
@@ -137,13 +138,14 @@ pub fn buildProviderConfig(
             .base_url = prov.base_url,
             .model = wire_model,
             .reasoning = reasoning,
+            .max_tokens = max_tokens,
         } },
         .anthropic_messages => return .{ .anthropic_messages = .{
             .api_key = prov.api_key,
             .base_url = prov.base_url,
             .model = wire_model,
             .api_version = if (def_opt) |d| (d.api_version orelse "2023-06-01") else "2023-06-01",
-            .max_tokens = if (def_opt) |d| (d.max_tokens orelse 4096) else 4096,
+            .max_tokens = max_tokens,
         } },
     }
 }
diff --git a/src/main.zig b/src/main.zig
index 20686eb..28f1144 100644
--- a/src/main.zig
+++ b/src/main.zig
@@ -704,6 +704,13 @@ fn persistTurn(
     model: []const u8,
     per_message_usage: []const ?panto.session_manager.Usage,
 ) !void {
+    var batch_messages: std.ArrayList(panto.session_manager.DiskMessage) = .empty;
+    defer batch_messages.deinit(alloc);
+    var batch_providers: std.ArrayList(?[]const u8) = .empty;
+    defer batch_providers.deinit(alloc);
+    var batch_models: std.ArrayList(?[]const u8) = .empty;
+    defer batch_models.deinit(alloc);
+
     var i = start_index;
     var assistant_seen: usize = 0;
     while (i < conv.messages.items.len) : (i += 1) {
@@ -718,22 +725,23 @@ fn persistTurn(
             blocks[allocated] = try panto.session.contentBlockToDisk(alloc, block);
             allocated += 1;
         }
+        if (msg.role == .assistant and hasToolUseWithoutFollowingResults(conv, i)) {
+            for (blocks[0..allocated]) |b| b.deinit(alloc);
+            alloc.free(blocks);
+            continue;
+        }
         switch (msg.role) {
             .system => {
                 // Mid-turn system messages aren't a thing in pantograph today.
                 // Treat as harmless and persist verbatim, sans stamps.
-                _ = try mgr.appendMessage(
-                    .{ .role = .system, .content = blocks },
-                    null,
-                    null,
-                );
+                try batch_messages.append(alloc, .{ .role = .system, .content = blocks });
+                try batch_providers.append(alloc, null);
+                try batch_models.append(alloc, null);
             },
             .user => {
-                _ = try mgr.appendMessage(
-                    .{ .role = .user, .content = blocks },
-                    provider,
-                    model,
-                );
+                try batch_messages.append(alloc, .{ .role = .user, .content = blocks });
+                try batch_providers.append(alloc, provider);
+                try batch_models.append(alloc, model);
             },
             .assistant => {
                 // Pair this assistant message with its usage, if the
@@ -745,19 +753,37 @@ fn persistTurn(
                 else
                     null;
                 assistant_seen += 1;
-                _ = try mgr.appendMessage(
-                    .{
-                        .role = .assistant,
-                        .content = blocks,
-                        .provider = try alloc.dupe(u8, provider),
-                        .model = try alloc.dupe(u8, model),
-                        .stop_reason = try alloc.dupe(u8, "stop"),
-                        .usage = usage,
-                    },
-                    null,
-                    null,
-                );
+                try batch_messages.append(alloc, .{
+                    .role = .assistant,
+                    .content = blocks,
+                    .provider = try alloc.dupe(u8, provider),
+                    .model = try alloc.dupe(u8, model),
+                    .stop_reason = try alloc.dupe(u8, "stop"),
+                    .usage = usage,
+                });
+                try batch_providers.append(alloc, null);
+                try batch_models.append(alloc, null);
             },
         }
     }
+    try mgr.appendMessagesAtomic(batch_messages.items, batch_providers.items, batch_models.items);
+}
+
+fn hasToolUseWithoutFollowingResults(conv: *const panto.conversation.Conversation, index: usize) bool {
+    const msg = conv.messages.items[index];
+    for (msg.content.items) |block| {
+        if (block != .ToolUse) continue;
+        if (index + 1 >= conv.messages.items.len) return true;
+        const next = conv.messages.items[index + 1];
+        if (next.role != .user) return true;
+        var found = false;
+        for (next.content.items) |next_block| {
+            if (next_block == .ToolResult and std.mem.eql(u8, next_block.ToolResult.tool_use_id, block.ToolUse.id)) {
+                found = true;
+                break;
+            }
+        }
+        if (!found) return true;
+    }
+    return false;
 }
diff --git a/src/models_toml.zig b/src/models_toml.zig
index 36d8c65..790b61f 100644
--- a/src/models_toml.zig
+++ b/src/models_toml.zig
@@ -10,7 +10,7 @@
 //!     [<provider>.<alias>]
 //!     model       = <string>  # wire model id; defaults to <alias> if omitted
 //!     reasoning   = <string>  # default | off | minimal | low | medium | high
-//!     max_tokens  = <int>     # anthropic_messages only
+//!     max_tokens  = <int>     # per-request output token cap
 //!     api_version = <string>  # anthropic_messages only
 //!     # pricing (all optional; USD per million tokens):
 //!     input       = <float>
@@ -64,7 +64,7 @@ pub const ModelDef = struct {
     /// TOML omitted `model`.
     model: []const u8,
     reasoning: ReasoningEffort,
-    /// anthropic_messages only; null = use the provider/library default.
+    /// Per-request output token cap; null = use the provider/library default.
     max_tokens: ?u32,
     /// anthropic_messages only; null = use the library default.
     api_version: ?[]const u8,