惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Google DeepMind News
Google DeepMind News
I
InfoQ
Engineering at Meta
Engineering at Meta
D
DataBreaches.Net
L
LangChain Blog
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Recent Announcements
Recent Announcements
GbyAI
GbyAI
爱范儿
爱范儿
Microsoft Security Blog
Microsoft Security Blog
腾讯CDC
美团技术团队
罗磊的独立博客
Microsoft Azure Blog
Microsoft Azure Blog
WordPress大学
WordPress大学
T
The Blog of Author Tim Ferriss
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
雷峰网
雷峰网
M
MIT News - Artificial intelligence
D
Docker
MongoDB | Blog
MongoDB | Blog
F
Fortinet All Blogs
博客园 - 叶小钗

Recent Commits to openclaw:main

test: merge chat side-result checks · openclaw/openclaw@ddd2c2a test: merge cron history checks · openclaw/openclaw@f7eb746 test: merge responsive navigation shell checks · openclaw/openclaw@c2e4b47 docs(changelog): add codex oauth fixes · openclaw/openclaw@628e6cd test: merge navigation routing cases · openclaw/openclaw@5d8cecb Tests: mock channel registry bundled fallback · openclaw/openclaw@2b08233 Secrets: avoid broad web search discovery for single plugin config · openclaw/openclaw@a464f59 test: merge config view browser checks · openclaw/openclaw@20cf511 fix(status): align oauth health with runtime · openclaw/openclaw@eed7116 feat: add macOS screen snapshots for monitor preview (#67954) thanks … · openclaw/openclaw@f377db1 fix: report shared auth scopes in hello-ok (#67810) thanks @BunsDev · openclaw/openclaw@0b6c39b Auto-reply: avoid eager bundled route fallback · openclaw/openclaw@3ea1bf4 Tests: narrow session binding contract setup · openclaw/openclaw@54e4e16 fix(macOS): enable undo/redo in webchat composer text input (#34962) · openclaw/openclaw@00951dc Tests: speed up channel setup promotion · openclaw/openclaw@82b529a Docs: refresh agent instructions · openclaw/openclaw@5775fe2 fix(auth): serialize OAuth refresh across agents to fix #26322 (#67876) · openclaw/openclaw@8e79080 test: allow ollama public surface boundary test · openclaw/openclaw@7d4f1a6 Docs: add test performance guardrails · openclaw/openclaw@89706d3 Tests: restore context-engine usage proof · openclaw/openclaw@e4c4f95 Tests: slim context engine runtime coverage · openclaw/openclaw@74c198f ci: retry failed custom checkouts · openclaw/openclaw@0ee5baf test: trim duplicate provider auth onboarding cases · openclaw/openclaw@1ffc02e matrix: fix sessions_spawn --thread subagent session spawning (#67643) · openclaw/openclaw@1ce2596 test: reduce auth choice fixture churn · openclaw/openclaw@857b9cd test: mock health status config boundaries · openclaw/openclaw@9d5ab4a test: mock onboard config io boundary · openclaw/openclaw@299694d test: mock legacy state plugin boundaries · openclaw/openclaw@2713089 test: mock channel install boundaries · openclaw/openclaw@b945248 test: mock doctor preview channel boundaries · openclaw/openclaw@b1a3ad4
fix(agents): prevent Bedrock replay death loop on empty a...
openperf · 2026-04-26 · via Recent Commits to openclaw:main

@@ -0,0 +1,115 @@

1+

import type { AgentMessage } from "@mariozechner/pi-agent-core";

2+

import { describe, expect, it } from "vitest";

3+

import { normalizeAssistantReplayContent } from "./replay-history.js";

4+5+

const FALLBACK_TEXT = "[assistant turn failed before producing content]";

6+7+

function bedrockAssistant(

8+

content: unknown,

9+

stopReason: "error" | "stop" | "toolUse" | "length" = "error",

10+

): AgentMessage {

11+

return {

12+

role: "assistant",

13+

content,

14+

api: "bedrock-converse-stream",

15+

provider: "amazon-bedrock",

16+

model: "anthropic.claude-3-haiku-20240307-v1:0",

17+

usage: {

18+

input: 0,

19+

output: 0,

20+

cacheRead: 0,

21+

cacheWrite: 0,

22+

totalTokens: 0,

23+

cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },

24+

},

25+

stopReason,

26+

timestamp: 0,

27+

} as unknown as AgentMessage;

28+

}

29+30+

function userMessage(text: string): AgentMessage {

31+

return { role: "user", content: text, timestamp: 0 } as unknown as AgentMessage;

32+

}

33+34+

function openclawTranscriptAssistant(model: "delivery-mirror" | "gateway-injected"): AgentMessage {

35+

return {

36+

role: "assistant",

37+

content: [{ type: "text", text: "channel mirror" }],

38+

api: "openai-responses",

39+

provider: "openclaw",

40+

model,

41+

usage: {

42+

input: 0,

43+

output: 0,

44+

cacheRead: 0,

45+

cacheWrite: 0,

46+

totalTokens: 0,

47+

cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },

48+

},

49+

stopReason: "stop",

50+

timestamp: 0,

51+

} as unknown as AgentMessage;

52+

}

53+54+

describe("normalizeAssistantReplayContent", () => {

55+

it("converts assistant content: [] to a non-empty sentinel text block when stopReason is error", () => {

56+

const messages = [userMessage("hello"), bedrockAssistant([], "error")];

57+

const out = normalizeAssistantReplayContent(messages);

58+

expect(out).not.toBe(messages);

59+

const repaired = out[1] as AgentMessage & { content: { type: string; text: string }[] };

60+

expect(repaired.content).toEqual([{ type: "text", text: FALLBACK_TEXT }]);

61+

});

62+63+

it("preserves silent-reply turns (stopReason=stop, content=[]) untouched", () => {

64+

// run.empty-error-retry.test.ts treats `stopReason:"stop"` + `content:[]`

65+

// as a legitimate NO_REPLY / silent-reply, NOT a crash. Substituting the

66+

// failure sentinel here would inject a fabricated "[assistant turn failed

67+

// before producing content]" into the next provider request and change

68+

// model behavior even though no failure occurred.

69+

const silentStop = bedrockAssistant([], "stop");

70+

const messages = [userMessage("hello"), silentStop];

71+

const out = normalizeAssistantReplayContent(messages);

72+

expect(out).toBe(messages);

73+

expect(out[1]).toBe(silentStop);

74+

});

75+76+

it("preserves empty content with non-error stopReasons (toolUse, length) untouched", () => {

77+

// Boundary lock: only `stopReason:"error"` should trip the sentinel

78+

// substitution. `toolUse` and `length` are reachable in practice when a

79+

// provider terminates a turn before a content block is emitted, and

80+

// rewriting them as a failure would lie about what happened.

81+

const toolUse = bedrockAssistant([], "toolUse");

82+

const length = bedrockAssistant([], "length");

83+

const messages = [userMessage("hello"), toolUse, length];

84+

const out = normalizeAssistantReplayContent(messages);

85+

expect(out).toBe(messages);

86+

expect(out[1]).toBe(toolUse);

87+

expect(out[2]).toBe(length);

88+

});

89+90+

it("wraps legacy string assistant content as a single text block (regression)", () => {

91+

const messages = [userMessage("hi"), bedrockAssistant("plain string content")];

92+

const out = normalizeAssistantReplayContent(messages);

93+

const wrapped = out[1] as AgentMessage & { content: { type: string; text: string }[] };

94+

expect(wrapped.content).toEqual([{ type: "text", text: "plain string content" }]);

95+

});

96+97+

it("filters openclaw delivery-mirror and gateway-injected assistant messages from replay", () => {

98+

const messages = [

99+

userMessage("hello"),

100+

openclawTranscriptAssistant("delivery-mirror"),

101+

bedrockAssistant([{ type: "text", text: "real reply" }]),

102+

openclawTranscriptAssistant("gateway-injected"),

103+

];

104+

const out = normalizeAssistantReplayContent(messages);

105+

expect(out).toHaveLength(2);

106+

expect((out[0] as { role: string }).role).toBe("user");

107+

expect((out[1] as { provider: string }).provider).toBe("amazon-bedrock");

108+

});

109+110+

it("returns the original array reference when nothing needs to change", () => {

111+

const messages = [userMessage("hello"), bedrockAssistant([{ type: "text", text: "fine" }])];

112+

const out = normalizeAssistantReplayContent(messages);

113+

expect(out).toBe(messages);

114+

});

115+

});