fix: emit a tool call that arrives with no arguments

A tool call whose arguments never arrive — a no-argument tool, or a model that
sends id and name and then stops — was accumulated into toolCallsState but
never added to the message, because the content block was only created inside
`if (tc.function?.arguments)`.

The finalizer then checked `toolCallsState.length > 0` and set
stopReason "toolUse" on a message containing no tool call at all. pi's agent
loop continues only on tool calls, so it had nothing to execute and the turn
ended: mid-task, no tool run, no error, nothing in the transcript except the
assistant saying what it was about to do. Observed against a real session where
the model announced "let me consult the advisor" four times in a row and the
run then stopped dead; its final message on disk is stopReason "toolUse" with a
single text block and no toolCall.

Two changes:

  - open the block as soon as the call is identifiable (id or name), and keep
    its id/name in step if they arrive in a later delta, since the finalizer
    only rewrites `arguments`;
  - guard the stopReason on blocks that actually reached the message rather
    than on the state array being non-empty, so a genuinely malformed stream
    reports "stop" instead of an unactionable "toolUse".

Tests: no-argument call, identity arriving after the block is open, and a
malformed delta that must not claim toolUse.
This commit is contained in:
Oyster Lee
2026-08-03 06:50:04 +08:00
parent a9a3521b9b
commit b4416909f5
2 changed files with 101 additions and 9 deletions
+34 -9
View File
@@ -407,17 +407,39 @@ export function streamQoder(
const state = toolCallsState[idx];
if (tc.id) state.id = tc.id;
if (tc.function?.name) state.name = tc.function.name;
// Open the block as soon as the call is IDENTIFIABLE, not
// when its first argument byte arrives. A call whose
// arguments are absent or an empty string — a no-argument
// tool, or a model that sends id+name and then stops — used
// to create a toolCallsState entry and no content block, so
// the finalizer below saw a non-empty state array, set
// stopReason "toolUse", and handed back a message with no
// tool call in it. The agent loop then had nothing to run
// and the turn simply ended, mid-task and without an error.
if (state.emittedStart === undefined && (state.id || state.name)) {
state.emittedStart = true;
state.contentIndex = output.content.length;
output.content.push({
type: "toolCall",
id: state.id,
name: state.name,
arguments: {},
} satisfies ToolCall);
stream.push({ type: "toolcall_start", contentIndex: state.contentIndex, partial: output });
}
// id and name can arrive after the block is open; keep it
// in step, since the finalizer only rewrites `arguments`.
if (state.emittedStart) {
const block = output.content[state.contentIndex] as ToolCall;
block.id = state.id;
block.name = state.name;
}
if (tc.function?.arguments) {
const argDelta = tc.function.arguments;
state.arguments += argDelta;
if (state.emittedStart === undefined) {
state.emittedStart = true;
state.contentIndex = output.content.length;
const block: ToolCall = { type: "toolCall", id: state.id, name: state.name, arguments: {} };
output.content.push(block);
stream.push({ type: "toolcall_start", contentIndex: state.contentIndex, partial: output });
}
stream.push({
type: "toolcall_delta",
contentIndex: state.contentIndex,
@@ -487,7 +509,10 @@ export function streamQoder(
}
}
if (toolCallsState.length > 0) {
// Guarded on blocks that actually reached the message, not on the state
// array being non-empty. Claiming "toolUse" for a message carrying no
// tool call is what turned a malformed stream into a silent dead end.
if (toolCallsState.some((state) => state?.emittedStart)) {
output.stopReason = "toolUse";
}
// Otherwise keep whatever finish_reason set upstream (defaults to "stop").