用debug执行例子就失败，不用debug就成功，是什么问题？

[anstarword]

autoqa run specs/saucedemo-01-login.md --url https://www.saucedemo.com/ --debug
runId=8ff49892-0262-45c8-ad5d-f8f09e7fb1ca
baseUrl=https://www.saucedemo.com
headless=false
debug=true
artifactRoot=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca
node=v22.14.0
parsedSpecs=1
spec=D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md
preconditions=3
steps=6
{"level":30,"timestamp":"2026-01-12T06:53:57.624Z","event":"autoqa.run.started","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","baseUrl":"https://www.saucedemo.com","headless":false,"debug":true,"artifactRoot":".autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","specCount":1}
{"level":30,"timestamp":"2026-01-12T06:53:58.018Z","event":"autoqa.spec.started","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","specPath":"D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md"}
mcp=browser (navigate/click/fill/scroll/wait)
system=init
mcp_status=[{"name":"browser","status":"connected"}]
I'll help you execute this AutoQA test for the SauceDemo login flow. Let me start by following the steps in order.

Step 1: Navigate to /

tool_use=mcp__browser__navigate id=toolu_018bYwDQeJRU8zA8Jpn97eJJ input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
tool_result id=toolu_018bYwDQeJRU8zA8Jpn97eJJ is_error=true content=Stream closed
Let me retry the navigation:
tool_use=mcp__browser__navigate id=toolu_0166rXxKz7pD6fNXbeY4hMEm input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
tool_result id=toolu_0166rXxKz7pD6fNXbeY4hMEm is_error=true content=Stream closed
tool_use=mcp__browser__navigate id=toolu_012u7NeLvf8cDQtLB8JNbxMu input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
tool_result id=toolu_012u7NeLvf8cDQtLB8JNbxMu is_error=true content=Stream closed
tool_use=mcp__browser__navigate id=toolu_01PXszoqpVGd6mwNpNny98ag input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
tool_result id=toolu_01PXszoqpVGd6mwNpNny98ag is_error=true content=Stream closed
tool_use=mcp__browser__navigate id=toolu_01GgGFCTWB8fH8j1juah2ANe input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
tool_result id=toolu_01GgGFCTWB8fH8j1juah2ANe is_error=true content=Stream closed
tool_use=mcp__browser__navigate id=toolu_01U1ZxsHJtYaqbQPhPbMTiYM input={"stepIndex":1,"url":"https://www.saucedemo.com/"}
{"level":30,"timestamp":"2026-01-12T07:06:15.710Z","event":"autoqa.guardrail.triggered","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","specPath":"D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md","stepIndex":1,"code":"GUARDRAIL_MAX_RETRIES_PER_STEP","limit":5,"actual":6}
{"level":30,"timestamp":"2026-01-12T07:06:15.810Z","event":"autoqa.spec.failure_screenshot","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","specPath":"D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md","screenshotPath":".autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/screenshots/failure-000-specs-saucedemo-01-login.jpg"}
{"level":30,"timestamp":"2026-01-12T07:06:15.871Z","event":"autoqa.spec.finished","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","specPath":"D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md","durationMs":737853,"ok":false,"failureReason":"SPEC_EXECUTION_FAILED: (GUARDRAIL_MAX_RETRIES_PER_STEP): GUARDRAIL_MAX_RETRIES_PER_STEP: limit=5 actual=6 stepIndex=1","tracePath":".autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/traces/000-specs-saucedemo-01-login.zip"}
{"level":30,"timestamp":"2026-01-12T07:06:16.148Z","event":"autoqa.run.finished","runId":"8ff49892-0262-45c8-ad5d-f8f09e7fb1ca","exitCode":1,"durationMs":738524,"specsPassed":0,"specsFailed":1,"failureSummary":"Failed to run spec: specs\saucedemo-01-login.md (GUARDRAIL_MAX_RETRIES_PER_STEP): GUARDRAIL_MAX_RETRIES_PER_STEP: limit=5 actual=6 stepIndex=1"}
specsPassed=0
specsFailed=1
durationMs=738524
logPath=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/run.log.jsonl
snapshotDir=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/snapshots
failedSpecPath=specs\saucedemo-01-login.md
screenshotsDir=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/screenshots
screenshotPath=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/screenshots/failure-000-specs-saucedemo-01-login.jpg
traceDir=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/traces
tracePath=.autoqa/runs/8ff49892-0262-45c8-ad5d-f8f09e7fb1ca/traces/000-specs-saucedemo-01-login.zip
failureSummary=Failed to run spec: specs\saucedemo-01-login.md (GUARDRAIL_MAX_RETRIES_PER_STEP): GUARDRAIL_MAX_RETRIES_PER_STEP: limit=5 actual=6 stepIndex=1
Failed to run spec: specs\saucedemo-01-login.md (GUARDRAIL_MAX_RETRIES_PER_STEP): GUARDRAIL_MAX_RETRIES_PER_STEP: limit=5 actual=6 stepIndex=1

autoqa run specs/saucedemo-01-login.md --url https://www.saucedemo.com/
runId=6ded1947-cffa-47c2-8b8a-a155d3f24ce6
baseUrl=https://www.saucedemo.com
headless=true
debug=false
artifactRoot=.autoqa/runs/6ded1947-cffa-47c2-8b8a-a155d3f24ce6
specsPassed=1
specsFailed=0
durationMs=45042
D:\ONE\TraeProjects\AutoQA\specs\saucedemo-01-login.md

[anstarword]

感觉是不稳定，我反复运行了多次autoqa run specs/saucedemo-01-login.md --url https://www.saucedemo.com/ 和 autoqa run specs/saucedemo-01-login.md --url https://www.saucedemo.com/ --debug
有时带debug可以pass，有时不带debug可以pass
(specsPassed=1)

[terryso]

大模型运行用例, 和你用的大模型和一些其它因素有关, 同一个用例的确不是每次都能成功的(这是目前大模型执行用例, 抽盲盒的一个通病)
所以我这个项目的目标是希望通过执行成功的那次, 将执行步骤导出成playwright的ts代码保存下来, 以后是直接运行playwright的ts代码, 而不是每次都用大模型执行用例.

如果每次都用大模型执行用例, 其实我这个项目就没必要存在, 直接用Claude code安装playwright mcp来执行自然语言用例就好

[anstarword]

理解：）
另外测试场景md文件里的参数化没太看懂，感觉是AUTOQA开头的变量可以在步骤中引入，但似乎没有正确传递到步骤中。如图：

大模型运行用例, 和你用的大模型和一些其它因素有关, 同一个用例的确不是每次都能成功的(这是目前大模型执行用例, 抽盲盒的一个通病) 所以我这个项目的目标是希望通过执行成功的那次, 将执行步骤导出成playwright的ts代码保存下来, 以后是直接运行playwright的ts代码, 而不是每次都用大模型执行用例.

如果每次都用大模型执行用例, 其实我这个项目就没必要存在, 直接用Claude code安装playwright mcp来执行自然语言用例就好

[terryso]

日志不会将环境变量中的值打印出来的, 尤其是PASSWORD, 你主要看一下执行用例的使用, 界面上输入的值是不是环境变量中的值

[anstarword]

日志会打出来的，我用你例子里默认的AUTOQA_USERNAME 和 AUTOQA_PASSWORD 以及 {{USERNAME}}，{{PASSWORD}}的时候，就可以正常使用变量传递，且会在日志里打印出值。所以没太搞懂这里面的逻辑。

[terryso]

你看agent在浏览器输入的值是不是环境变量中的值, 还是也是没有输入环境变量中的值

[anstarword]

有时候是环境变量，有时不是