合规闸门
Apothem 合规闸门——每个发布的工件都必须通过的十五道预发布闸门关卡(M1-M15)中的机械部分。
十五道关卡
| 关卡 | ID | 强制规定 | 检查类别 | 失败 → 行动 |
|---|---|---|---|---|
| 1 | M1 | 宿主项目无关性 | 推理判断 | 修订以匹配宿主约定 |
| 2 | M2 | 编辑纪律(披露账本) | 推理判断 | 添加披露标记 |
| 3 | M3 | 十项质量维度 | 机械 + 推理判断 | 针对未通过的维度进行修订 |
| 4 | M4 | 自我应用(存在验证块) | 推理判断 | 记录验证块 |
| 5 | M5 | 权威原则(不捏造身份 / 端点 / 机密) | 机械 | 转交询问 |
| 6 | M6 | 专业知识融入(已揭示的缺口) | 推理判断 | 揭示缺口;引用改进之处 |
| 7 | M7 | 选项标注(推荐标记 + 驱动因素理由) | 机械 | 标注选项集 |
| 8 | M8 | 确定性(规定性文字中不含含糊措辞) | 机械 | 将含糊措辞提升为条件句 |
| 9 | M9 | 视觉杠杆(结构性主题配有图示) | 机械 + 推理判断 | 编写 / 刷新图示 |
| 10 | M10 | 双向绑定(互逆的反向指针已闭合) | 机械 | 闭合半边 |
| 11 | M11 | 敏捷冲刺(目标 / 待办列表 / 就绪定义 / 完成定义 / 评审 / 回顾) | 推理判断 | 重构为冲刺 |
| 12 | M12 | 阶段报告与规范布局 | 机械 + 推理判断 | 编写汇总;迁移到规范布局 |
| 13 | M13 | 代码工艺(lint / 格式化 / 类型检查;魔法数字;裸 except) | 机械 | 应用各语言的代码工艺规则 |
| 14 | M14 | 系统性(声明上游 / 下游 / 同级 / 执行者) | 机械 + 推理判断 | 声明;在登记册中编入索引 |
| 15 | M15 | 生产就绪(测试 + 文档 + CHANGELOG + 合规提交 + CI 绿色 + semver 稳定性) | 机械 + 推理判断 | 补齐缺失的表面 |
闸门只运行可由机器检查的校验器。推理判断类关卡所要求的推理判断,以及每个机械 + 推理判断类关卡中的推理判断部分,都在评审中评估。
严重级别阶梯
闸门没有关键、重要或建议性的分级,也没有两项发现的阈值。不带 --strict(或 APOTHEM_CONFORMITY_STRICT=1)时,闸门只报告发现项,并以 0 退出。带 --strict 时,单文件或钩子检查在任一匹配器失败时以 2 退出;语料库运行(--all 或 --all-perwrite)在某个阻断性校验器或匹配器失败时以 2 退出;建议性发现项只报告。已安装的 Write/Edit 钩子不传递 --strict,因此它只报告发现项,写入会继续进行,除非其环境中设置了 APOTHEM_CONFORMITY_STRICT=1。
校验器
机械部分位于 src/apothem/conformity/,由 src/apothem/conformity/gate.py 编排。每个校验器以下列三种方式之一运行:
- per-write:在每次待处理的写入时运行(
--hook),或用--all-perwrite对每个已跟踪的文件运行。 - standalone:用
--all在整个仓库范围内运行,或用--check <name>逐个运行。 - change-set:针对已暂存的差异,直接以
python -m apothem.conformity.semver_stability_grep --staged运行,不经由闸门。
下表由校验器模块生成。
The conformity package ships 54 validators: 22 per-write, 31 standalone and 1 change-set. The gate registers the per-write and standalone ones.
| Validator | Runs | Rule anchor | What it checks |
|---|---|---|---|
agent-capability-grep | standalone | agent-capability-discipline (per-harness capability matrix) | Verify every harness declares its agentic-capability matrix. |
agents-md-coverage-grep | standalone | rules/agents-md-convention.md | Flag stale agent-companion files under the root-only AGENTS.md convention. |
agnosticism-grep | standalone | rules/agnostic-posture.md §3 (harness neutrality) + §1 (default-off) | Flag harness bias and re-introduced enforcement presets in shipped surfaces. |
always-on-budget-grep | per-write | D1 token-budget-discipline §Always-on body budget | Body-budget conformity matcher for always-on rules. Pointed at a rules directory it also holds the loaded always-on set and the rule index to byte ceilings. |
bare-except-grep | per-write | M13.3 error handling | Flag bare except clauses without typed exceptions or re-raise. |
binding-five-direction-grep | standalone | M10 bidirectional-binding §1 (five-direction Bindings section) | Require the five-direction Bindings section on every rule, command, agent, and skill. |
binding-reciprocity-corpus-grep | standalone | M10 bidirectional-binding §2 (↔ self-reciprocal) | Cross-file binding-reciprocity corpus walk — the ↔-symmetric half-edge check. |
binding-reciprocity-grep | per-write | M10 bidirectional-binding §Notation | Flag binding declarations using non-canonical arrow notation. |
brand-mark-grep | per-write | rules/plain-language.md (brand-mark display/identifier casing) | Flag brand-mark display/identifier form drift on brand-touched surfaces. |
commented-out-code-grep | per-write | M13.1 why-not-what comments | Flag substantial commented-out code blocks left in source. |
completion-claim-grep | per-write | rules/session-closure.md (verifiable-close output honesty) | Flag unsupported completion / guarantee claims in cohort prose. |
conventional-commit-grep | standalone | CLAUDE.md Coding Conventions — Conventional Commits | Flag HEAD commit messages that drift from the Conventional-Commits grammar. |
copilot-instructions-presence-grep | per-write | — | GitHub Copilot surface presence check. |
cross-platform-matrix-grep | standalone | cross-platform OS x Python CI matrix | Verify CI declares a cross-platform OS x Python matrix. |
determinism-grep | standalone | rules/determinism.md | Prove every command and skill surface has a deterministic output shape. |
diagram-staleness-grep | per-write | M9 visual-leverage §Fidelity | Flag diagrams whose verification date is missing or stale. |
dynamism-grep | standalone | rules/dynamism.md (static-version embeds) | Flag static-string version embeds in dynamism-required surfaces. |
editorconfig-presence-grep | standalone | supply-chain SOTA-rigor §editorconfig | Verify a canonical .editorconfig is present at the project root. |
file-header-grep | per-write | — | Authorship-header presence and canonical-form check. |
freshness-token-grep | standalone | rules/freshness-facade.md | Flag freshness-narrative phrases on shipped public surfaces. |
frontmatter-grep | per-write | the per-class frontmatter JSON Schemas (src/apothem/schemas/) | Validate per-artifact-class YAML frontmatter against required-key sets. |
frontmatter-value-grep | standalone | per-class JSON Schema enum/pattern value constraints | Validate frontmatter VALUES against the per-class JSON Schema enum/pattern constraints. |
gate-evidence-grep | standalone | commands/plan-execute.md Gate evidence; rules/pre-emission-gate-bars.md section 2 (attestation schema) | Advisory: flag plan-suite gate claims that lack a command, an exit status or output, and a HEAD SHA beside them, and fifteen-bar attestations that omit a field of the rule's schema. |
gitattributes-presence-grep | standalone | supply-chain .gitattributes contract | Verify the repo-root .gitattributes carries the canonical contract. |
harden-runner-grep | standalone | harden-runner egress baseline | Verify every workflow job opens with a conformant harden-runner step. |
hedging-grep | per-write | M8 definitiveness §Hedging | Flag hedging vocabulary in prescriptive contexts. |
license-author-consistency-grep | per-write | — | LICENSE author-identity presence check. |
link-check | per-write | rules/code-craft-markdown.md §6 Link Discipline | Validate Markdown internal links against the on-disk reference graph. |
magic-number-grep | per-write | M13.10 magic numbers | Flag magic numbers in source code. |
multi-surface-coherence-grep | per-write | — | Shared-claim coherence across AI-conventions surfaces. |
naming-grep | standalone | CLAUDE.md Coding Conventions (kebab-case naming) | Validate filesystem paths against the canonical naming convention. |
no-global-plans-grep | standalone | CLAUDE.md Plans Discipline (no global plans) | Walk the git index for tracked plans paths; fail when found. |
no-toplevel-docs-grep | standalone | single-source-of-truth (docs live in site/, not root docs/) | Fail when a top-level docs/ directory exists at the repository root. |
oidc-trusted-publishing-grep | standalone | OIDC trusted publishing | Verify release workflows use OIDC trusted publishing. |
option-annotation-grep | per-write | M7 option-annotation (H6 per-option bind) | Enforce the per-option Recommended label-to-body bind (H6). |
orphan-output-grep | per-write | M12 canonical-layout + M14 systemicity | Flag artifacts emitted to non-canonical locations or without provenance. |
permissions-minimum-scope-grep | standalone | minimum-scope workflow permissions | Verify every GitHub Actions workflow declares a minimum-scope permissions: block. |
plain-language-grep | standalone | rules/plain-language.md — user-facing prose | Flag mechanistic / harness-internal vocabulary in user-facing prose. |
plan-next-step-consistency-grep | standalone | rules/recommend-next-step.md | Advisory: flag plan-suite infra files whose Recommended-Next-Step footer drifts from the suite's recorded next pipeline stage. |
plan-suite-structure-grep | standalone | context-management-scratch 2 (suite-locality + closed vocab) + canonical-layout-reporting-tiers 2.1 (numeric-prefix discipline) | Validate plan-suite structure: suite-locality, closed vocab, numeric prefix. |
plans-discipline-language-grep | standalone | CLAUDE.md Plans Discipline (multi-surface language) | Verify the canonical Plans Discipline language reaches every required surface. |
production-ready-pr-grep | per-write | M15 production-ready §Same-change-set | Flag commits whose change-set lacks the production-ready four-class shape. |
recommend-next-step-grep | standalone | rules/recommend-next-step.md | Verify every command file carries a ## Recommended Next Step block. |
redundancy-grep | standalone | rules/operational-mandates.md §CM-7 coherent-product non-redundancy | Flag substantively-duplicated paragraphs across the governed corpus. |
reference-token-grep | standalone | rules/own-voice-reimplementation.md | Flag leaked reference-platform branding tokens on authored surfaces. |
registry-capability-consistency-grep | standalone | rules/agent-capability-discipline.md — registry capability cells backed by evidence | Flag registry capability cells not backed by install / materializer / projection evidence. |
secret-leak-grep | per-write | M13.8 security + M15 production-ready | Block emission of artifacts containing hardcoded credentials. |
semver-stability-grep | change-set | M15 semver-stability | Semver stability matcher (M15 mechanical fraction). |
smoke-install-grep | standalone | production-ready section 8.1 paired install scripts | Smoke-test the paired install scripts at the repository root. |
static-version-grep | standalone | rules/dynamism.md §static-version-resolution | Verify version-bearing sites resolve dynamically, not as static literals. |
token-efficiency-grep | per-write | clean-room-generation §5 + token-efficiency-rewrite | Flag filler phrases, throat-clearing openers, and content-free qualifiers. |
unpinned-action-grep | per-write | M15 production-ready §Supply-chain | Flag GitHub Actions workflows that pin actions to mutable refs. |
user-confirm-grep | per-write | M5 authority-inquiry §10 | Block emission of artifacts containing unresolved authority placeholders. |
workflow-concurrency-grep | standalone | workflow concurrency + per-job timeouts | Enforce concurrency + per-job timeout discipline on GitHub workflows. |
运行闸门
在已安装的工具上,闸门会自动运行:Claude Code 的 settings.json 将其接入为一个 PreToolUse 钩子,在每次 Write/Edit 之前调用位于 ~/.claude/.apothem/support/conformity/gate.py 的物化副本。随后闸门只检查其钩子范围内的写入,范围之外的写入会无声通过;该范围列于运行合规闸门。在源码检出中,模块形式驱动同一套代码;请在下面每条 python 命令前加上 PYTHONPATH=src。闸门默认是建议性的:它报告发现项并以 0 退出。添加 --strict 可在出现阻断性发现项时以 2 退出。未知的校验器名称以 3 退出。
# Every standalone validator across the repository:
python -m apothem.conformity.gate --all .
# The same sweep, failing on a blocking finding (the CI form):
python -m apothem.conformity.gate --all --strict .
# Every per-write validator against every tracked file:
python -m apothem.conformity.gate --all-perwrite .
# One validator by name (names come from --list):
python -m apothem.conformity.gate --check naming-grep .
# List every registered validator as JSON:
python -m apothem.conformity.gate --list
# Single-file dispatch (mimics the PreToolUse hook):
echo '{"tool_input": {"file_path": "path/to/file", "content": "..."}}' \
| python -m apothem.conformity.gate --hookpython -m apothem.conformity.gate --help 会打印用法摘要。完整流程参见运行合规闸门。