dtcg-tokens
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinesedtcg-tokens
dtcg-tokens
Turn any inspectable multimodal input into source-specific DTCG 2025.10 JSON, evidence JSON, a standalone HTML proof artifact, and a run record. Start with the full possibility surface, then narrow it with located source evidence and reasons. Strong native vision owns source interpretation, design decisions, artifact authorship, and final visual judgment. Never invent a source, audience, product fact, brand premise, comparison corpus, or proof claim.
将任何可检查的多模态输入转换为源特定的DTCG 2025.10 JSON、证据JSON、独立HTML证明工件以及运行记录。从完整的可能性范围入手，然后通过定位到的源证据和理由缩小范围。强大的原生视觉能力负责源解读、设计决策、工件创作以及最终视觉判断。切勿虚构源、受众、产品事实、品牌前提、对比语料库或证明声明。
Commands
命令
| Command | Result |
|---|---|
| Show inputs, intents, outputs, vision requirements, and claim limits. |
| Create |
| Check DTCG 2025.10 structure, values, names, references, type agreement, and cycles. |
| Author, assemble, inspect, and read back a source- and token-specific standalone proof artifact. |
Load , , , or only for the matching command. Load only for an absolute uniqueness request. Load , , and only in Step 24.
examples/help.mdexamples/generate.mdexamples/validate.mdexamples/prove.mdexamples/failure-global-claim.mdevals/contract.mdevals/rubric.mdscripts/tests/test_ci_contract.py| 命令 | 结果 |
|---|---|
| 显示输入、意图、输出、视觉要求以及声明限制。 |
| 创建 |
| 检查DTCG 2025.10的结构、值、名称、引用、类型一致性以及循环。 |
| 创作、组装、检查并回读特定于源和令牌的独立证明工件。 |
仅在匹配对应命令时加载、、或。仅在请求绝对唯一性时加载。仅在第24步加载、和。
examples/help.mdexamples/generate.mdexamples/validate.mdexamples/prove.mdexamples/failure-global-claim.mdevals/contract.mdevals/rubric.mdscripts/tests/test_ci_contract.pyPackage map
包映射
A directory name does not load a file. Load, inspect, copy, or run a file only when the current command, Execution row, step procedure, or Completion gate names its exact relative path.
| Path | Contains | Load or run when |
|---|---|---|
| Plain-language procedures, rules, decision guides, and review methods. | Read only the exact Markdown file routed by the current step or gate. |
| Machine-readable schemas, catalogs, contracts, policies, manifests, templates, and static fixtures. | Inspect the named asset when a procedure or script consumes it; never treat the directory as instruction text. |
| Versioned creative primitives, operators, mechanisms, concepts, themes, constraints, questions, technology records, and negative patterns. | Load through its manifest only during the exploration step or its validation. |
| Deterministic command-line scaffolds, preparers, assemblers, validators, linters, and audits. | Run the exact script named by the current step or gate; read its code only to repair or audit it. |
| Import-only helpers for DTCG rules, accounting, coverage, artifact checks, review, fonts, experiments, and pipeline state. | Load only through the script that imports the named helper; never run a library file as a command. |
| Discovered unit and contract tests for script and package behavior. | Run through |
| Trigger, behavior, failure, recovery, lineage, mapping, contract, and rubric records. | Load only the named eval file during Step 24 or a targeted evaluation. |
| Bounded token, evidence, source, and negative visual fixtures used by named evals or tests. | Load only the fixture named by its eval, mapping, or test; never treat a rejected fixture as a positive precedent. |
| Verified worked command runs and one bounded failure example. | Read only the example that matches the active command or failure. |
| The local task graph and the single full-check entry point. | Run or inspect it in Step 24 and package maintenance. |
| Remote checks that call the same local task entry point. | Inspect only when auditing remote validation or repository integration. |
目录名称不会加载文件。仅当当前命令、执行行、步骤流程或完成门限指定其确切相对路径时，才加载、检查、复制或运行文件。
| 路径 | 包含内容 | 加载或运行时机 |
|---|---|---|
| 通俗易懂的流程、规则、决策指南和评审方法。 | 仅读取当前步骤或门限指向的具体Markdown文件。 |
| 机器可读的模式、目录、契约、策略、清单、模板和静态固定数据。 | 当流程或脚本需要使用指定资源时检查该资源；切勿将目录视为指令文本。 |
| 带版本的创意原语、操作符、机制、概念、主题、约束、问题、技术记录和负面模式。 | 仅在探索步骤或其验证阶段通过其清单加载。 |
| 确定性命令行脚手架、准备工具、组装工具、验证器、代码检查器和审计工具。 | 运行当前步骤或门限指定的具体脚本；仅在修复或审计时读取其代码。 |
| 用于DTCG规则、统计、覆盖检查、工件验证、评审、字体、实验和流水线状态的仅导入型辅助工具。 | 仅通过导入该辅助工具的脚本加载；切勿将库文件作为命令运行。 |
| 针对脚本和包行为的已发现单元测试与契约测试。 | 通过 |
| 触发记录、行为记录、失败记录、恢复记录、lineage记录、映射记录、契约记录和评分标准记录。 | 仅在第24步或针对性评估时加载指定的评估文件。 |
| 由指定评估或测试使用的受限令牌、证据、源和负面视觉固定数据。 | 仅加载其评估、映射或测试指定的固定数据；切勿将被拒绝的固定数据视为正面先例。 |
| 已验证的命令运行示例和一个受限失败示例。 | 仅读取与当前活跃命令或失败场景匹配的示例。 |
| 本地任务图和唯一的全检查入口点。 | 在第24步和包维护阶段运行或检查。 |
| 调用相同本地任务入口点的远程检查。 | 仅在审计远程验证或仓库集成时检查。 |
Operating contract
操作契约
Fix the process, not the design. Keep step order, named handoffs, evidence, states, gates, and claim limits deterministic. Keep source types, intents, extensions, token content, visual language, specimens, interactions, qualitative lenses, and experiments open. Use whenever a catalog omits a supported input or idea. For typography, follow and run ; each selected family must rank outside the most popular 50% of the same-date live Google Fonts catalog and pass source-specific vision review.
references/extension-protocol.mdreferences/google-font-selection.mdscripts/prepare_google_fonts.pyPush every mechanical action into when it can be defined by inputs, outputs, schemas, hashes, state transitions, comparisons, counts, or exit codes. Scripts scaffold judgment; they never make the judgment. Use to initialize the run, hash inputs, prepare each step packet, enforce prerequisites, record outputs, and set or . A strong vision-capable model must own every visual or qualitative decision. If the active executor lacks strong vision, delegate every judgment and do not generate tokens or proof until that route passes .
scripts/scripts/run_pipeline.pyPASSBLOCKEDreferences/vision-execution.md修复流程，而非设计。保持步骤顺序、指定的交接、证据、状态、门限和声明限制的确定性。保持源类型、意图、扩展、令牌内容、视觉语言、样本、交互、定性视角和实验的开放性。当目录中缺少支持的输入或想法时，使用。排版方面，遵循并运行；每个选定的字体家族必须在同期实时Google Fonts目录的流行度排名中处于后50%，并通过源特定的视觉评审。
references/extension-protocol.mdreferences/google-font-selection.mdscripts/prepare_google_fonts.py当机械操作可由输入、输出、模式、哈希、状态转换、比较、计数或退出码定义时，将其全部移入。脚本为判断提供框架；但绝不做出判断。使用初始化运行、哈希输入、准备每个步骤数据包、强制执行先决条件、记录输出并设置或。具备强大视觉能力的模型必须负责所有视觉或定性决策。如果当前执行器缺乏强大视觉能力，则委托所有判断，直到该流程通过后再生成令牌或证明。
scripts/scripts/run_pipeline.pyPASSBLOCKEDreferences/vision-execution.mdExecution
执行流程
generateprovehelpvalidatescripts/run_pipeline.py packetscripts/run_pipeline.py startpassblockassets/execution-io-map.jsonassets/execution-step-contract.jsonreferences/deterministic-execution.mdInputActionSavePassBlockedFeeds| # | Step | Use | Produces |
|---|---|---|---|
| 01 | Freeze the run | | |
| 02 | Inventory input and intent | | |
| 03 | Prove vision execution | | |
| 04 | Observe visuals | | |
| 05 | Classify statements | | |
| 06 | Falsify the thesis | | |
| 07 | Expand the token universe | | |
| 08 | Generate experiments | | |
| 09 | Map contexts | | |
| 10 | Create signatures | | |
| 11 | Author tokens | | |
| 12 | Validate tokens | | |
| 13 | Build evidence | | |
| 14 | Author proof | | |
| 15 | Reject reused shells | | |
| 16 | Expose coverage | | |
| 17 | Assemble proof | | |
| 18 | Inspect final HTML | | |
| 19 | Resolve defects | | |
| 20 | Resolve invariants | | |
| 21 | Judge four tracks | | |
| 22 | Repair causes | | |
| 23 | Read back final bytes | | |
| 24 | Validate the package | | |
| 25 | Return proved results | | |
generateprovehelpvalidatescripts/run_pipeline.py packetscripts/run_pipeline.py startpassblockassets/execution-io-map.jsonassets/execution-step-contract.jsonreferences/deterministic-execution.mdInputActionSavePassBlockedFeeds| 序号 | 步骤 | 使用工具 | 产出物 |
|---|---|---|---|
| 01 | 冻结运行 | | |
| 02 | 盘点输入与意图 | | |
| 03 | 证明视觉执行能力 | | |
| 04 | 观察视觉内容 | | |
| 05 | 分类陈述 | | |
| 06 | 证伪论点 | | |
| 07 | 扩展令牌范围 | | |
| 08 | 生成实验 | | |
| 09 | 映射上下文 | | |
| 10 | 创建签名 | | |
| 11 | 创作令牌 | | |
| 12 | 验证令牌 | | |
| 13 | 构建证据 | | |
| 14 | 创作证明 | | |
| 15 | 拒绝复用框架 | | |
| 16 | 展示覆盖范围 | | |
| 17 | 组装证明 | | |
| 18 | 检查最终HTML | | |
| 19 | 解决缺陷 | | |
| 20 | 解决不变量 | | |
| 21 | 评审四个维度 | | |
| 22 | 修复根源 | | |
| 23 | 回读最终字节 | | |
| 24 | 验证包 | | |
| 25 | 返回已验证结果 | | |
Boundaries
边界限制
- Token count is not quality evidence. Unsupported tokens count against the result.
- Keep conformance, source specificity, taste, originality, corpus uniqueness, non-AI-slop, invariants, and artifact integrity as separate claims.
- Token JSON cannot prove rendered quality. Machine checks cannot certify pixels. Vision cannot override measured failures.
- Experimental tokens may remain without a passing use claim. The proof must expose failures and limits as clearly as passes.
- Use for taste, originality, corpus uniqueness, and non-AI-slop judgment; keep its interpretation flexible and its evidence requirements fixed.
references/qualitative-judgment.md - Global uniqueness is outside this skill's valid claim scope and remains unproved.
- 令牌数量并非质量证据。不支持的令牌会对结果产生负面影响。
- 将一致性、源特异性、品味、原创性、语料库唯一性、非AI垃圾内容、不变量和工件完整性作为独立声明。
- 令牌JSON无法证明渲染质量。机器检查无法认证像素效果。视觉判断无法覆盖已测量的失败。
- 实验性令牌可无需通过使用声明保留。证明必须清晰展示失败和限制，如同展示成功一样。
- 使用进行品味、原创性、语料库唯一性和非AI垃圾内容的判断；保持其解释灵活性，同时固定其证据要求。
references/qualitative-judgment.md - 全局唯一性不在本技能的有效声明范围内，且无法被证明。
Completion
完成条件
| Gate | Done only when | Check with |
|---|---|---|
| Pipeline | All 25 current step records are | |
| Deliverables | | |
| Conformance | | |
| Accounting | Every source, intent, catalog leaf, screen possibility, context cell, extension, exclusion, and retained token path appears exactly once. | |
| Exploration | At least three retained experimental tokens use at least three distinct strategies, include inversion or antithesis, and agree across tokens, evidence, specimens, and final review. | |
| Specificity | At least five source-derived signature decisions across at least three applicable axes trace from located source evidence to token paths and rendered regions. | |
| Typography | Same-date live Google Fonts research yields at least three eligible candidates, each selection ranks outside the most popular 50%, embedded WOFF2 and license hashes match, offline rendering works, and vision review passes. | |
| Proof coverage | Every applicable token type, context permutation, and experimental token has a meaning-specific visible specimen; raw JSON alone does not pass. | |
| Visual readback | The final assembled | |
| Defects and invariants | Every defect and invariant has one disposition, with zero unresolved vetoes, zero unresolved major defects, and no failed claimed invariant. | |
| Judgment and claims | Taste, originality, corpus uniqueness, and non-AI-slop are judged separately with located evidence, counterevidence, uncertainty, competing readings, and limits; | |
| Package health | Every persistent file has a named activation or consumer route, every required check uses current bytes, and fresh | |
If any gate fails or lacks evidence, use under ; return , name each failed gate and exact recovery need, and withhold non-slop, originality, corpus uniqueness, and completion claims.
scripts/run_pipeline.py blockreferences/deterministic-execution.mdBLOCKED| 门限 | 仅在满足以下条件时完成 | 检查工具 |
|---|---|---|
| 流水线 | 所有25个当前步骤记录均为 | |
| 交付物 | | |
| 一致性 | | |
| 统计 | 每个源、意图、目录节点、屏幕可能性、上下文单元、扩展、排除和保留的令牌路径均恰好出现一次。 | |
| 探索 | 至少三个保留的实验性令牌使用至少三种不同策略，包含反转或对立论点，且在令牌、证据、样本和最终评审中保持一致。 | |
| 特异性 | 至少五个源自源的签名决策（覆盖至少三个适用维度）可从定位到的源证据追溯到令牌路径和渲染区域。 | |
| 排版 | 同期实时Google Fonts研究产生至少三个合格候选字体，每个选定字体的流行度排名处于后50%，嵌入的WOFF2和许可证哈希匹配，离线渲染可用，且通过视觉评审。 | |
| 证明覆盖范围 | 每个适用的令牌类型、上下文排列和实验性令牌都有一个特定含义的可见样本；仅原始JSON无法通过。 | |
| 视觉回读 | 最终组装的 | |
| 缺陷与不变量 | 每个缺陷和不变量都有一个处理结果，无未解决的否决项、无未解决的重大缺陷，且无已声明的不变量失败。 | |
| 判断与声明 | 品味、原创性、语料库唯一性和非AI垃圾内容已通过定位到的证据、反证、不确定性、竞争性解读和限制分别进行判断； | |
| 包健康 | 每个持久文件都有指定的激活或消费路径，每个必要检查都使用当前字节，且新运行的 | |
如果任何门限失败或缺少证据，请根据运行；返回，命名每个失败的门限和确切的修复需求，并保留非垃圾内容、原创性、语料库唯一性和完成声明。
references/deterministic-execution.mdscripts/run_pipeline.py blockBLOCKED