[WIP] refactor/cleanup repo

[cquil11]

Repository refactoring progress

Tracks merged work related to repository cleanup, shared Python tooling, workflow simplification, and recipe organization.

Merged: core refactoring

• refactor: deduplicate benchmark internals / 重构：去除基准测试内部重复逻辑 #2897 — Deduplicate benchmark internals.

• refactor: simplify changelog orchestration and strengthen tests / 简化变更日志编排并强化测试 #2901 — Simplify changelog orchestration and strengthen tests.

• refactor: share AgentX matrix generation / 共用 AgentX 矩阵生成逻辑 #2903 — Share AgentX matrix generation.

• refactor: move matrix tooling into infx / 将矩阵工具迁移到 infx 包 #2911 — Move matrix tooling and schemas into infx.

• refactor: centralize results in infx and speed up tests / 集中结果处理并加速测试 #2940 — Centralize result processing in infx and speed up tests.

• refactor: move power engines into infx / 将功耗引擎迁入 infx #2941 — Move power engines into infx.

• refactor: share eval artifact selection rules / 共享评测工件选择规则 #2947 — Share eval artifact selection rules.

• refactor: build eval rows in infx / 将评测行构建迁移至 infx #2948 — Move eval row construction into infx.

• refactor: share result topology through infx / 通过 infx 共享结果拓扑逻辑 #2987 — Share result topology and GPU-count logic.

• refactor: move AgentX processing into infx / 将 AgentX 处理迁入 infx #3003 — Move AgentX processing into infx.

• refactor: retain eval winners during reuse deduplication / 复用去重时保留已选评测结果 #3015 — Retain selected eval results during reuse deduplication.

• refactor: centralize eval result selection in infx / 将评测结果选择集中到 infx #3018 — Centralize eval result selection in infx.

• refactor: plan sweeps in process through infx / 通过 infx 在进程内规划扫描 #3019 — Plan sweeps in-process through infx.

• refactor: migrate remaining Python tools into infx / 将剩余 Python 工具迁入 infx #3022 — Migrate remaining Python tools into infx.

• refactor: unify matrix expansion / 统一矩阵展开逻辑 #3029 — Unify matrix expansion.

• refactor: standardize srt-slurm v2.2.1 and migrate recipes #3086 — Standardize srt-slurm v2.2.1 and shared launcher integration.

• refactor: unify benchmark workflow inputs / 统一基准工作流输入 #3094 — Unify benchmark workflow inputs.

• refactor: migrate and organize srt-slurm recipes #3103 — Migrate and organize SRT recipes. Stacked on and integrated into refactor: standardize srt-slurm v2.2.1 and migrate recipes #3086.

• Package infx and centralize uv dependencies / 打包 infx 并统一管理 uv 依赖 #3168 — Package infx, centralize locked uv dependencies with a 12-hour cutoff, and resolve trusted tooling from current main.

• Unify GitHub requests through gh and validate listings / 统一通过 gh 请求 GitHub 并验证列表 #3180 — Unify GitHub requests through gh api and reject malformed or incomplete listings while preserving workflow tokens and local authentication.

• refactor: share eval artifact loading and validation / 重构：统一评测工件读取与校验 #3193 — Share eval artifact loading and validation across collection, reuse, and Klaud while preserving selection and failure ordering.

• Use native SRT overrides with automatic golden acceptance #3152 — Replace manual synthetic-acceptance settings and the YAML injector with native SRT overrides and automatic golden-curve selection; includes documented acceptance corrections.

Merged: tests, CI, and cleanup

• test: remove tautological tests and document test quality #2799 — Remove tautological tests and document test-quality expectations.

• refactor(klaud): concise guidance and progress history / 精简指导并记录进度历史 #2898 — Consolidate Klaud guidance and progress reporting.

• ci: adopt uv and harden workflow role checks / 迁移 CI 至 uv 并加强 Workflow 角色检查 #2943 — Standardize CI on uv and harden workflow role checks.

• [PowerX] audit power tests and remove the one tautological file, keeping B200 exporter coverage / 审计功耗测试，删除唯一重言文件并保留 B200 覆盖 #3089 — Remove redundant power tests while retaining exporter coverage.

• docs: collapse Chinese translations / 折叠中文翻译 #3093 — Collapse Chinese translations in GitHub templates and guidance.

• test: prune structural, tautological, and duplicate tests #3101 — Prune structural, tautological, and duplicate tests.

• fix(ci): reuse checklist comment and retain the first PR approval #3125 — Reuse one checklist verdict comment; carry-forward policy refined in fix: harden pipeline edge cases and retain sign-off across admin updates / 修复流水线边界问题并在管理员更新后保留签核 #3145.

• fix: update one PR comment with the latest sweep visualizer links #3126 — Update one sweep visualizer comment and prevent stale runs from replacing newer links.

• ci: add ruff linting and formatting for infx #3140 — Add Ruff linting and formatting for infx.

• Clean up obsolete tests, eval aliases, and runtime patches #3141 — Remove obsolete tests, eval aliases, and runtime patches; centralize runtime settings.

• docs: drop CodeQL from the recover-failed-ingest check list #3143 — Remove obsolete CodeQL guidance from ingest recovery instructions.

• fix: harden pipeline edge cases and retain sign-off across admin updates / 修复流水线边界问题并在管理员更新后保留签核 #3145 — Harden pipeline edge cases, trim weak tests, and retain sign-off across verified admin updates.

• chore: strengthen Ruff checks for infx / 加强 infx 的 Ruff 检查 #3153 — Strengthen Ruff checks and clean up infx implementations.

• fix: skip sign-off verification without performance changelog changes #3154 — Skip unnecessary sign-off runs; superseded by ownership-based scoping in fix: consolidate CI and scope CODEOWNER sign-off / 合并 CI 并按文件归属判断 CODEOWNER 签核 #3155.

• fix: consolidate CI and scope CODEOWNER sign-off / 合并 CI 并按文件归属判断 CODEOWNER 签核 #3155 — Consolidate CI workflows and scope CODEOWNER sign-off.

• fix: preserve Klaud baselines and fill candidate slots / 保留完整基线并补足候选 #3157 — Preserve complete Klaud baselines and fill candidate slots past duplicates or capacity claims.

• ci: enforce strict zizmor audits and harden workflows / 启用严格 zizmor 审计并加固工作流 #3161 — Enforce strict Zizmor checks for workflow changes and harden workflow execution.

• fix: handle duplicate paths in sign-off checks / 修复签核检查对重复路径的处理 #3162 — Handle duplicate GitHub file entries without weakening incomplete-diff checks.

• fix: use current target-branch CODEOWNERS / 使用目标分支最新 CODEOWNERS 修复签核 #3163 — Read current target-branch CODEOWNERS so existing PRs receive ownership fixes.

• fix: clarify sign-off verdict output / 精简签核结论并明确拒绝状态 #3164 — Simplify sign-off verdict output and distinguish rejection from missing verification.

• Pin E2E benchmarks to the matrix checkout / 将 E2E 基准测试固定到矩阵 checkout 提交 #3177 — Pin E2E benchmark and eval jobs to the resolved measured commit. Stacked on and integrated into Package infx and centralize uv dependencies / 打包 infx 并统一管理 uv 依赖 #3168.

• Fail closed when collecting workflow statistics / 运行统计收集异常时明确失败 #3179 — Fail closed on workflow-statistics collection errors, share GitHub pagination, and remove PyGithub.

• Add /use as the preferred sweep reuse command / 新增首选扫描复用命令 /use #3192 — Add /use <run_id> as the preferred sweep-reuse command, preserve legacy syntax, and protect unrelated comment reactions.

• docs: clarify advisory CODEOWNER verification / 文档：说明 CODEOWNER 自动验证仅供参考 #3197 — Document advisory CODEOWNER verification after removing its required-status gate; preserve checklist feedback and human review requirements.

• refactor: simplify advisory CODEOWNER verification / 重构：简化 CODEOWNER 参考验证 #3199 — Remove advisory sign-off status publishing and carry-forward bookkeeping; retain checklist verification and one comment tied to the assessed commit.

• refactor: remove unused Klaud shims / 重构：删除未使用的 Klaud 兼容入口 #3201 — Delete unused utils/klaud compatibility wrappers; current callers already use infx.klaud.

• refactor: share sweep lookup with staging / 重构：统一暂存与扫描复用查询 #3202 — Share sweep-run lookup between staging and reuse, move staging into infx, preserve command behavior, and consolidate behavioral tests.

• ci: reduce PAT usage and simplify secret names / 减少 PAT 使用并简化密钥命名 #3212 — Scope workflow credentials by purpose, use the built-in token for checkout and run statistics, and retain the existing Hugging Face secret.

• fix: enable Klaud fail-fast final sweeps / 支持 Klaud 最终完整扫描快速失败 #3208 — Support Klaud fail-fast final sweeps while retaining exact-head, successful-run, and complete-coverage validation.

• fix(klaud): keep baseline only in PR body / 将 Klaud 基线仅保留在 PR 正文 #3233 — Keep the complete frozen Klaud baseline in the PR body, remove duplicate baseline comments, and retain recovery of the old format.

中文

本清单记录已合并的仓库重构、共享 Python 工具、工作流简化和配方整理工作。

本次补充的已合并工作包括：折叠 GitHub 模板与指南中的中文翻译；复用单一签核结论和 sweep 可视化评论，防止旧运行覆盖新链接；清理过时的 CodeQL 恢复指南；修复流水线边界情况、删减低质量测试，并在已验证的管理员更新后保留签核。

签核范围检查的后续修复包括：正确处理 GitHub 返回的重复文件记录，同时拒绝不完整的差异列表；读取目标分支最新的 CODEOWNERS，使已有 PR 也能获得归属修复；精简结论输出，明确区分验证被拒绝与尚需验证。早期的签核延续和跳过规则已由清单中注明的后续 PR 调整。

新增的已合并工作：#3157 保留完整的 Klaud 基线，并在遇到重复候选或容量占用时继续补齐候选名额；#3161 为工作流变更启用严格的 Zizmor 检查并加固执行边界；#3168 将 infx 打包，使用 uv 统一锁定依赖并实施 12 小时发布冷却期，可信工具统一取自当前 main；#3177 固定 E2E 基准测试与评估使用的测量提交，已随 #3168 集成。

#3179 修复运行统计在请求失败时仍报告成功的问题，统一使用共享 GitHub 分页逻辑，并移除 PyGithub 及不再使用的依赖。

#3180 统一通过 gh api 发送 GitHub 请求，拒绝格式错误或不完整的列表，同时保留工作流 Token 和本地认证方式。

#3192 新增首选扫描复用命令 /use <run_id>，保留旧命令，并保护无关评论上的表态。

#3193 统一收集、复用校验和 Klaud 的评测工件读取与校验，保留结果选择规则和错误处理顺序。

#3197 说明 CODEOWNER 自动验证在移除必需状态门禁后仅供参考，保留清单反馈与人工审阅要求。

#3199 移除参考签核的状态发布及批准延续逻辑，保留清单验证和注明实际评估提交的单一评论。

#3201 删除未使用的 utils/klaud 兼容入口，现有调用均已直接使用 infx.klaud。

#3202 统一暂存与扫描复用的源运行查询，将暂存逻辑移入 infx，保留命令行为并精简行为测试。

#3212 按用途隔离工作流凭据，checkout 和运行统计改用内置令牌，并保留现有 Hugging Face secret。

#3152 使用原生 SRT overrides 和自动 golden 曲线选择，替代手动 synthetic acceptance 设置与 YAML 注入器；同时包含 PR 中明确说明的 acceptance 修正。

#3208 支持 Klaud 的 fail-fast 最终完整 sweep，同时保留精确 head、运行成功和完整覆盖校验。

#3233 将完整冻结基线仅保留在 Klaud PR 正文中，移除重复的基线评论，并兼容旧格式的恢复流程。