Skip to content

feat(ISSUE-263): W3-C3 llm-verifier 接入 (#279) - #72

Merged
cloudbrid-agent[bot] merged 1 commit into
mainfrom
feat/ISSUE-263-w3-c3
Aug 23, 2026
Merged

feat(ISSUE-263): W3-C3 llm-verifier 接入 (#279)#72
cloudbrid-agent[bot] merged 1 commit into
mainfrom
feat/ISSUE-263-w3-c3

Conversation

@cloudbrid-agent

Copy link
Copy Markdown
Contributor

摘要

接入 llm-verifier(pip 钉点+lock):criteria 一卡一文件且机器可追溯到卡 AC;endpoint 三探测每次 run 前执行(精度折损声明);token 账挂接 ADR-0062 metering + usage 交叉核对(偏差超阈值 run 作废转人工)。

变更

  • pipeline/adversary/llm_verifier.py:endpoint 三探测(logprobs/top_logprobs/structured_outputs,不满足 fail-closed)+ K 次重复评估 + 阈值 gate + token 交叉核对
  • pipeline/adversary/verifier-config.yaml:verifier 角色档与采样参数锁定(与 adversary-config.yaml 同构)
  • pipeline/adversary/criteria/ISSUE-263/ac-criteria.yaml:示例 criteria 文件(AC-1/AC-2/AC-10/AC-11)
  • scripts/llm-verify.sh:CLI 入口
  • .github/requirements-llm-verifier.txt:pip 钉点 pyyaml==6.0.3

接口兼容

与 W3-C1/W3-C5 接口兼容(输入输出 JSON schema 一致,verifier-report/v1)。

Closes Cloudbird-Software/.github#279

接入 llm-verifier(pip 钉点+lock):
- pipeline/adversary/llm_verifier.py:endpoint 三探测(logprobs/top_logprobs/
  structured_outputs,fail-closed)+ K 次重复评估 + criteria 逐文件溯源 AC +
  token 账与 metering 交叉核对(偏差超阈值作废转人工)
- pipeline/adversary/verifier-config.yaml:verifier 角色档与采样参数锁定
- pipeline/adversary/criteria/ISSUE-263/ac-criteria.yaml:示例 criteria 文件
- scripts/llm-verify.sh:CLI 入口(封装 llm_verifier.py)
- .github/requirements-llm-verifier.txt:pip 钉点 pyyaml==6.0.3

与 W3-C1/W3-C5 接口兼容(输入输出 JSON schema 一致)。

Closes Cloudbird-Software/.github#279
@cloudbrid-agent
cloudbrid-agent Bot enabled auto-merge (squash) August 23, 2026 08:43
@cloudbrid-agent
cloudbrid-agent Bot merged commit a5f4ea3 into main Aug 23, 2026
21 checks passed
@cloudbrid-agent
cloudbrid-agent Bot deleted the feat/ISSUE-263-w3-c3 branch August 23, 2026 08:43
randypanding added a commit that referenced this pull request Aug 26, 2026
对近一周(#21..#124)全部 PR 复盘后的机械债清理:仅删除 AST 级验证
「全仓零引用」的未用导入/未用名,不改任何判定逻辑、阈值、白名单或
policy 数据。逐文件出处:

- pipeline/adversary/cnb_bridge.py:删未用 `from typing import Any`(#73/#74)
- pipeline/adversary/golden_set.py:删未用 `from typing import Any`(#80/#82/#83)
- pipeline/adversary/holdout_registry.py:删未用 `from typing import Any`(#81/#82)
- pipeline/adversary/e2e/e2e-runner.py:删未用 `from typing import Any`(#89)
- pipeline/adversary/llm_verifier.py:删未用 `import math`;可选库导入行去掉
  未用名 extract_score(call_verifier/create_openai_client 均在用,保留)(#72/#76)
- pipeline/entropy/tests/test_e2e.py:删未用 `import sys`(#56)
- pipeline/selftest-c/tests/test_registry.py:删未用 `import copy`(#103)
- pipeline/trust-gate/tests/test_adjudicate.py:删未用 `import copy`(#63)
- pipeline/trust-gate/tests/test_cli.py:from-import 去掉未用名
  PREDICATES/UNLOCK_STATE(保留 trust_gate 可导入性冒烟导入与 noqa 惯例)(#63)
- scripts/dep-supply-chain-check.py:删未用 `import copy`(#36/#43)

刻意不动(已核验非死代码):各模块 `from __future__ import annotations`;
fuzz/sast/symbolic 的 `_yamlmini` 双模式导入守卫(noqa F401,保证包路径);
golden_set 等 try-import yaml 的环境 fail-closed 守卫;org-gate /
suppression-gate / adversary-gate 等关卡 workflow 与 policy/suppressions.yaml
基线数据——门语义一概不变。

验证:
- py_compile 全部 scripts/pipeline *.py 通过;bash -n 全部 *.sh 通过
- workflows/policy/pipeline 共 62 个 YAML 解析通过
- scripts/test-integrity-fixtures/run.sh、scripts/suppression-budget-selftest.sh 通过
- python -m unittest:trust-gate test_adjudicate+test_cli 17 例、
  selftest-c tests.test_registry 14 例、entropy tests.test_e2e 10 例——全绿

Co-authored-by: randypanding <randypanding@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

W3-C3: llm-verifier 接入(criteria 派生 + 三探测 + 钉点 + token 账)

1 participant