Repository navigation
Conversation
…n a bot check fill_credential(name, field, selector) types a username, password or the current RFC 6238 code from a credential the spec names by reference (an env var or a file), only on the credential's own origin and through type_text's own gates. The value is in no argument, result, trace row or report; every password and seed is also scrubbed from what a run writes and returns, in plain, URL-encoded and JSON forms. navigate counts each bot check per vendor and site and answers challenge_abandoned at challenge_max_attempts (3); in a UAT run that ends the browsing and the report's exit reason. captcha_solver (module:function) is an operator hook for interactive checks when nobody watches; none ships, and the page is observed again rather than the solver believed. scripts/verify_unattended_e2e.py signs in on the demo site (password + TOTP) in real Chrome, meets the gate three times, and finds neither canary in any file of either run.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
무엇
credentials는 이름별로 비밀이 있는 위치(env 변수 또는 파일)만 적고, 비밀 자체는 적지 않는다. 페르소나는fill_credential(name, field, selector, submit)를 쓰고,field는username/password/totp다. 서버가 값을 해석해type_text의 본체를 거쳐 입력하므로 동의·가드·audit redaction 이 그대로 적용된다. 입력은 credential 의 origin 에서만 하고, 다른 곳에서는wrong_origin을 돌려주며 아무것도 입력하지 않는다. 비밀을 읽을 수 없으면credential_unavailable이다. TOTP 는 stdlib 로 구현했다(lyra_browser/totp.py, RFC 6238, 의존성 추가 없음).<secret>으로 바꾼다. 적용 대상은 trace/events 쓰기, 모델이 받는 도구 결과,network.jsonl(캡처된 POST body)이다.navigate의 challenge 응답에attempts·max_attempts를 붙인다. 같은 검사(vendor×origin)를LYRA_BROWSER_CHALLENGE_MAX_ATTEMPTS(기본 3) 번째로 만나면challenge_abandoned를 낸다. UAT run 에서는 이후 브라우저 도구를 모두 거부하고 policy event 를 남기며, exit reasonchallenge_abandoned(incomplete)로 끝난다. 이 exit reason 은 feat(uat): disposable profiles, retries, start limiter, SIGKILL sweeper #22 의 재시도 대상이 아니다.captcha_solver훅:LYRA_BROWSER_CAPTCHA_SOLVER/target.captcha_solver=module:function. 아무도 보지 않을 때 interactive 검사에만 호출한다. lyra 는 solver 를 하나도 싣지 않는다. solver 의 답은 믿지 않고 페이지를 다시 관찰해서, 검사가 사라졌을 때만ok+challenge_solved로 돌려준다.scripts/verify_unattended_e2e.py를 추가하고, 데모 사이트에/login→/2fa→/account를 더했다.왜
무인 실행에서 로그인, 2FA, 봇 검사를 처리해야 한다. 비밀은 모델, trace, 리포트 어디에도 남지 않아야 한다.
검증
pytest -q: 2002 passed, 5 failed. 실패 5건(test_uat_llm_brains.py)은 이 venv 에 SDK 가 없어서 나며, origin/master 에서도 똑같이 실패한다.test_totp.py: RFC 6238 부록 B 벡터 18개(sha1/256/512) 포함.test_challenge.py7개: 분기, 3회 포기, vendor 별 카운트, solver 성공/실패, attended·blocked 일 때 solver 미호출, 로더.tests/uat/test_uat_credentials.py5개: 실제 입력, origin·누락·미지 거부, run 디렉터리 전체에서 canary 0건, 페이지가 비밀을 echo 해도 scrub, challenge_abandoned 로 run 종료.scripts/verify_unattended_e2e.py, headless, 2026-11-15): native·playwright 모두 ALL PASS./account에 도달./gate3회에서 상태가 challenge, challenge, challenge_abandoned 순으로 나왔고, 이후 호출은 거부됨.<secret>이 들어 있음.남은 것 / 범위 밖
spec.py,recorder.py,server.py,docs/UAT.md를 같이 건드린다.