Skip to content

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

fix(agent): retain notice on empty response cutoff

Emre Sokullu9eb77710b56e
diff --git a/src/chrome/src/agent/agent.js b/src/chrome/src/agent/agent.js
index 5c92c9dc1..eeff1b25e 100644
--- a/src/chrome/src/agent/agent.js
+++ b/src/chrome/src/agent/agent.js
@@ -26277,7 +26277,17 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
// A reply that stopped at the output limit is an error on some providers. The
// side-panel answer still shows what was written, marked as cut off below.
const partial = phase === 'response_only' ? cutOffPartialFromError(error) : '';
- if (!partial) throw error;
+ if (!partial) {
+ // OpenAI Responses can spend the entire output budget on reasoning. If that
+ // same call exhausted the allowance, retain its notice rather than routing
+ // it through generic failure recovery, which does not preserve it.
+ if (phase === 'response_only'
+ && error?.costAllowanceMessage
+ && ['length', 'max_tokens', 'max_output_tokens', 'max_completion_tokens'].includes(String(error?.incompleteReason))) {
+ return `${RESPONSE_CUT_OFF_NOTICE.trim()}\n\n${error.costAllowanceMessage}`;
+ }
+ throw error;
+ }
result = {
content: Agent._stripReasoningTags(partial),
toolCalls: [],
diff --git a/src/firefox/src/agent/agent.js b/src/firefox/src/agent/agent.js
index e649937b4..134127e12 100644
--- a/src/firefox/src/agent/agent.js
+++ b/src/firefox/src/agent/agent.js
@@ -23418,7 +23418,17 @@ Rules: no prose intro, no conclusion, no "this screenshot shows...", no layout d
// A reply that stopped at the output limit is an error on some providers. The
// side-panel answer still shows what was written, marked as cut off below.
const partial = phase === 'response_only' ? cutOffPartialFromError(error) : '';
- if (!partial) throw error;
+ if (!partial) {
+ // OpenAI Responses can spend the entire output budget on reasoning. If that
+ // same call exhausted the allowance, retain its notice rather than routing
+ // it through generic failure recovery, which does not preserve it.
+ if (phase === 'response_only'
+ && error?.costAllowanceMessage
+ && ['length', 'max_tokens', 'max_output_tokens', 'max_completion_tokens'].includes(String(error?.incompleteReason))) {
+ return `${RESPONSE_CUT_OFF_NOTICE.trim()}\n\n${error.costAllowanceMessage}`;
+ }
+ throw error;
+ }
result = {
content: Agent._stripReasoningTags(partial),
toolCalls: [],
diff --git a/test/post-submit-model-recovery.mjs b/test/post-submit-model-recovery.mjs
index e55491463..ba9d7e6d4 100644
--- a/test/post-submit-model-recovery.mjs
+++ b/test/post-submit-model-recovery.mjs
@@ -153,6 +153,27 @@ for (const browser of ['chrome', 'firefox']) {
}
});
+ test(`${browser}: textless response-only cutoffs retain their allowance notice`, async () => {
+ const f = fixture(); f.agent._persist = () => {};
+ const notice = 'Cloud cost allowance reached: total cloud/router usage is $5.00 against the $5.00 limit.';
+ f.agent._completionEvidenceMessages = (_tab, messages) => messages;
+ f.agent._pruneOldImages = messages => messages;
+ f.agent._providerMaxOutputTokens = () => 100;
+ f.agent._chatWithCostAllowance = async () => {
+ throw Object.assign(new Error('Output limit reached'), {
+ incompleteReason: 'max_output_tokens',
+ partialContent: '',
+ costAllowanceMessage: notice,
+ });
+ };
+ const provider = { contextWindow: 128000 };
+ const answer = await f.agent._generateContextOnlyResponse(f.tab, f.messages, provider, {}, null, {
+ phase: 'response_only', step: 1,
+ });
+ assert.match(answer, /output limit and was cut off/i);
+ assert.match(answer, /cost allowance reached/i);
+ });
+
test(`${browser}: a stopped interaction offers Retry without replaying tools or overriding publication audit`, async () => {
const f = fixture(); f.agent._persist = () => {};
f.agent._generateContextOnlyResponse = async () => '';