asecli 0.8.0__tar.gz → 0.8.1__tar.gz

{asecli-0.8.0 → asecli-0.8.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: asecli
-Version: 0.8.0
+Version: 0.8.1
 Summary: Agent-native CLI for reading, modifying, validating and compiling Amplify Shader Editor assets
 Project-URL: Homepage, https://github.com/seeseeczl/ASECLI
 Project-URL: Repository, https://github.com/seeseeczl/ASECLI
@@ -28,7 +28,7 @@ Description-Content-Type: text/markdown
 
 AseCLI 是面向 AI Agent 的本地 CLI：离线读取、编辑和校验 ASE 节点图；通过运行中的 Unity/团结引擎与 MCP，让真实 ASE 创建节点、提供精排几何并重新生成 HLSL。Skill 提供跨 Agent 的操作规则，CLI 提供可验证的执行结果。
 
-当前正式版：**[0.8.0](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.0)** · [PyPI](https://pypi.org/project/asecli/0.8.0/) · [Agent 操作手册](skills/asecli/SKILL.md) · [SG 转换后整理流程](skills/asecli/references/conversion-workflow.md)
+当前正式版：**[0.8.1](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.1)** · [PyPI](https://pypi.org/project/asecli/0.8.1/) · [Agent 操作手册](skills/asecli/SKILL.md) · [SG 转换后整理流程](skills/asecli/references/conversion-workflow.md)
 
 ASECLI 接收 SGCLI 生成的裸 `EditorGraphSpec v3`，并可把受支持的 ASE Shader 直接导出为 SGCLI 的裸 `sgcli.native.v3`。两个方向都只有“生产者 spec 原文件→消费者 `--spec`”；独立 report 与发布 receipt 都不进入创建输入，创建成功也不代表转换效果等价。
 
@@ -132,7 +132,7 @@ curl -fsSL https://raw.githubusercontent.com/seeseeczl/ASECLI/main/scripts/insta
 uv tool uninstall asecli
 ```
 
-当前正式版本为 `0.8.0`（双向消费者原生裸 JSON 规格直连与可审计发布证据），可从 [PyPI](https://pypi.org/project/asecli/0.8.0/) 或 [GitHub Release](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.0) 安装。历史版本见 [GitHub Releases](https://github.com/seeseeczl/ASECLI/releases)，`v0.7.0` 保留为回滚点。
+当前正式版本为 `0.8.1`（透明 Multiply 的 matcher-v4 语义修复与转换认证加固），可从 [PyPI](https://pypi.org/project/asecli/0.8.1/) 或 [GitHub Release](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.1) 安装。历史版本见 [GitHub Releases](https://github.com/seeseeczl/ASECLI/releases)，`v0.8.0` 保留为回滚点。
 
 已知限制：真实 MCP 下，属性+纹理 v2 和 Reciprocal 降级 v2 的 create、manifest、validate、独立 recompile 已通过；SGCLI SphereMask v3 spec 在重载时仍可能返回 `node missing after reload: Radius` 并回滚。本次不承诺该 recipe 端到端可用，也不改变 SGCLI v2 默认值契约。
 
@@ -162,7 +162,7 @@ asecli --help
 
 ```bash
 shasum -a 256 -c SHA256SUMS
-uv tool install /path/to/asecli-0.8.0-py3-none-any.whl
+uv tool install /path/to/asecli-0.8.1-py3-none-any.whl
 asecli --help
 ```
 
@@ -207,7 +207,7 @@ sgcli sg create Assets/ConvertedNew.shadergraph \
 
 `export-sg` 自动从源 Shader 向上识别 Unity/Tuanjie 工程；跨工程资源解析时显式传 `--target-project`。它强制生成 `<stem>.asecli-to-sgcli.spec.json`、`.report.json` 与 `.receipt.json`，同名即失败；receipt 只证明 spec/report 两份文件的提交状态与 SHA，不是中间交接包。命令不调用 SGCLI，也不生成 ASE 自有中间 JSON。未知节点、Master/Pass、外部属性或纹理 Importer 语义时只写规范报告，不写半成品 spec。反方向 `.asecli-to-sgcli.spec.json` 不能交给 ASECLI `create --spec`。
 
-ASECLI 随 wheel 发布自己拥有的 EditorGraphSpec v3 Schema，并固定一份 SGCLI `0.3.0` 的 `sgcli.native.v3` 生产快照；SGCLI 同样固定 ASECLI `0.8.0` 快照。双仓兼容检查会比较两组原始 Schema SHA-256，任何漂移都必须先升级生产者再放行。
+ASECLI 随 wheel 发布自己拥有的 EditorGraphSpec v3 Schema，并固定一份 SGCLI `0.3.3` 的 `sgcli.native.v3` 生产快照；SGCLI 同样固定 ASECLI `0.8.1` 快照。双仓兼容检查会比较两组原始 Schema SHA-256，任何漂移都必须先升级生产者再放行。
 
 可用 `asecli contract editor-graph --version 3 --spec FILE` 在不连接 Editor 时执行权威 Schema/loader 并返回原文件 SHA。报告分开记录来源快照、发布前复核、生产者 Schema 与真实消费者读取状态。旧 `{graph,report}` wrapper 只允许按已登记形状显式迁移，且内层必须已经是 `sgcli.native.v3`；无版本 discriminator 的旧输入只报告 `legacy_format`，不得伪称已识别版本化 Schema：
 
@@ -340,9 +340,9 @@ asecli recompile Assets/Example.shader
 除显式 `--help` 保留人类可读文本外，stdout 恒为单行严格 JSON，agent 可直接解析。`--version` 也返回 JSON：
 
 ```json
-{"contract_version":1,"cli_version":"0.8.0","command":"parse","ok":true,"data":{"node_count":7}}
-{"contract_version":1,"cli_version":"0.8.0","command":"parse","ok":false,"error":{"code":"NOT_FOUND","message":"..."}}
-{"contract_version":1,"cli_version":"0.8.0","command":"version","ok":true,"data":{"version":"0.8.0"}}
+{"contract_version":1,"cli_version":"0.8.1","command":"parse","ok":true,"data":{"node_count":7}}
+{"contract_version":1,"cli_version":"0.8.1","command":"parse","ok":false,"error":{"code":"NOT_FOUND","message":"..."}}
+{"contract_version":1,"cli_version":"0.8.1","command":"version","ok":true,"data":{"version":"0.8.1"}}
 ```
 
 常见错误码：`PARSE_ERROR`、`NOT_FOUND`、`USAGE_ERROR`、`SCHEMA_UNAVAILABLE`、`SCHEMA_VERSION_MISMATCH`、`VALIDATION_ERROR`、`PROPERTY_PRESENTATION_ERROR`、`LAYOUT_ERROR`、`CHECKSUM_FORMAT_ERROR`、`GUI_SUPPORT_ERROR`、`CUSTOM_GUI_ERROR`、`COMMENT_GROUP_ERROR`、`GRAPH_REVIEW_ERROR`、`SEMANTIC_MISMATCH`、`SG_EXPORT_BLOCKED`、`EXTERNAL_REFERENCE`、`SKILL_INSTALL_CONFLICT`、`SKILL_INSTALL_ERROR`、`WRITE_CONFLICT`、`WRITE_PARTIAL`、`UNSAFE_PATH`、`WRITE_ERROR`、`BRIDGE_ERROR`、`INTERNAL`。

{asecli-0.8.0 → asecli-0.8.1}/README.md

@@ -4,7 +4,7 @@
 
 AseCLI 是面向 AI Agent 的本地 CLI：离线读取、编辑和校验 ASE 节点图；通过运行中的 Unity/团结引擎与 MCP，让真实 ASE 创建节点、提供精排几何并重新生成 HLSL。Skill 提供跨 Agent 的操作规则，CLI 提供可验证的执行结果。
 
-当前正式版：**[0.8.0](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.0)** · [PyPI](https://pypi.org/project/asecli/0.8.0/) · [Agent 操作手册](skills/asecli/SKILL.md) · [SG 转换后整理流程](skills/asecli/references/conversion-workflow.md)
+当前正式版：**[0.8.1](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.1)** · [PyPI](https://pypi.org/project/asecli/0.8.1/) · [Agent 操作手册](skills/asecli/SKILL.md) · [SG 转换后整理流程](skills/asecli/references/conversion-workflow.md)
 
 ASECLI 接收 SGCLI 生成的裸 `EditorGraphSpec v3`，并可把受支持的 ASE Shader 直接导出为 SGCLI 的裸 `sgcli.native.v3`。两个方向都只有“生产者 spec 原文件→消费者 `--spec`”；独立 report 与发布 receipt 都不进入创建输入，创建成功也不代表转换效果等价。
 
@@ -108,7 +108,7 @@ curl -fsSL https://raw.githubusercontent.com/seeseeczl/ASECLI/main/scripts/insta
 uv tool uninstall asecli
 ```
 
-当前正式版本为 `0.8.0`（双向消费者原生裸 JSON 规格直连与可审计发布证据），可从 [PyPI](https://pypi.org/project/asecli/0.8.0/) 或 [GitHub Release](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.0) 安装。历史版本见 [GitHub Releases](https://github.com/seeseeczl/ASECLI/releases)，`v0.7.0` 保留为回滚点。
+当前正式版本为 `0.8.1`（透明 Multiply 的 matcher-v4 语义修复与转换认证加固），可从 [PyPI](https://pypi.org/project/asecli/0.8.1/) 或 [GitHub Release](https://github.com/seeseeczl/ASECLI/releases/tag/v0.8.1) 安装。历史版本见 [GitHub Releases](https://github.com/seeseeczl/ASECLI/releases)，`v0.8.0` 保留为回滚点。
 
 已知限制：真实 MCP 下，属性+纹理 v2 和 Reciprocal 降级 v2 的 create、manifest、validate、独立 recompile 已通过；SGCLI SphereMask v3 spec 在重载时仍可能返回 `node missing after reload: Radius` 并回滚。本次不承诺该 recipe 端到端可用，也不改变 SGCLI v2 默认值契约。
 
@@ -138,7 +138,7 @@ asecli --help
 
 ```bash
 shasum -a 256 -c SHA256SUMS
-uv tool install /path/to/asecli-0.8.0-py3-none-any.whl
+uv tool install /path/to/asecli-0.8.1-py3-none-any.whl
 asecli --help
 ```
 
@@ -183,7 +183,7 @@ sgcli sg create Assets/ConvertedNew.shadergraph \
 
 `export-sg` 自动从源 Shader 向上识别 Unity/Tuanjie 工程；跨工程资源解析时显式传 `--target-project`。它强制生成 `<stem>.asecli-to-sgcli.spec.json`、`.report.json` 与 `.receipt.json`，同名即失败；receipt 只证明 spec/report 两份文件的提交状态与 SHA，不是中间交接包。命令不调用 SGCLI，也不生成 ASE 自有中间 JSON。未知节点、Master/Pass、外部属性或纹理 Importer 语义时只写规范报告，不写半成品 spec。反方向 `.asecli-to-sgcli.spec.json` 不能交给 ASECLI `create --spec`。
 
-ASECLI 随 wheel 发布自己拥有的 EditorGraphSpec v3 Schema，并固定一份 SGCLI `0.3.0` 的 `sgcli.native.v3` 生产快照；SGCLI 同样固定 ASECLI `0.8.0` 快照。双仓兼容检查会比较两组原始 Schema SHA-256，任何漂移都必须先升级生产者再放行。
+ASECLI 随 wheel 发布自己拥有的 EditorGraphSpec v3 Schema，并固定一份 SGCLI `0.3.3` 的 `sgcli.native.v3` 生产快照；SGCLI 同样固定 ASECLI `0.8.1` 快照。双仓兼容检查会比较两组原始 Schema SHA-256，任何漂移都必须先升级生产者再放行。
 
 可用 `asecli contract editor-graph --version 3 --spec FILE` 在不连接 Editor 时执行权威 Schema/loader 并返回原文件 SHA。报告分开记录来源快照、发布前复核、生产者 Schema 与真实消费者读取状态。旧 `{graph,report}` wrapper 只允许按已登记形状显式迁移，且内层必须已经是 `sgcli.native.v3`；无版本 discriminator 的旧输入只报告 `legacy_format`，不得伪称已识别版本化 Schema：
 
@@ -316,9 +316,9 @@ asecli recompile Assets/Example.shader
 除显式 `--help` 保留人类可读文本外，stdout 恒为单行严格 JSON，agent 可直接解析。`--version` 也返回 JSON：
 
 ```json
-{"contract_version":1,"cli_version":"0.8.0","command":"parse","ok":true,"data":{"node_count":7}}
-{"contract_version":1,"cli_version":"0.8.0","command":"parse","ok":false,"error":{"code":"NOT_FOUND","message":"..."}}
-{"contract_version":1,"cli_version":"0.8.0","command":"version","ok":true,"data":{"version":"0.8.0"}}
+{"contract_version":1,"cli_version":"0.8.1","command":"parse","ok":true,"data":{"node_count":7}}
+{"contract_version":1,"cli_version":"0.8.1","command":"parse","ok":false,"error":{"code":"NOT_FOUND","message":"..."}}
+{"contract_version":1,"cli_version":"0.8.1","command":"version","ok":true,"data":{"version":"0.8.1"}}
 ```
 
 常见错误码：`PARSE_ERROR`、`NOT_FOUND`、`USAGE_ERROR`、`SCHEMA_UNAVAILABLE`、`SCHEMA_VERSION_MISMATCH`、`VALIDATION_ERROR`、`PROPERTY_PRESENTATION_ERROR`、`LAYOUT_ERROR`、`CHECKSUM_FORMAT_ERROR`、`GUI_SUPPORT_ERROR`、`CUSTOM_GUI_ERROR`、`COMMENT_GROUP_ERROR`、`GRAPH_REVIEW_ERROR`、`SEMANTIC_MISMATCH`、`SG_EXPORT_BLOCKED`、`EXTERNAL_REFERENCE`、`SKILL_INSTALL_CONFLICT`、`SKILL_INSTALL_ERROR`、`WRITE_CONFLICT`、`WRITE_PARTIAL`、`UNSAFE_PATH`、`WRITE_ERROR`、`BRIDGE_ERROR`、`INTERNAL`。

{asecli-0.8.0 → asecli-0.8.1}/docs/00-governance/change-register.md

@@ -43,6 +43,6 @@ evidence: [docs/05-audits/2026-09-01-122217-optimization-tasks.md]
 | CR-0027 | 将随 wheel 分发的 ASECLI Skill 从 Codex 默认安装扩展为主流 Agent Skills 通用安装 | MOD-CLI MOD-SKILL | 新增 `--agent`、`--scope`、`--project-root`；开放 `.agents/skills` 覆盖 Codex/Cursor/Gemini CLI/GitHub Copilot，Claude Code 使用 `.claude/skills`；无参数与 `--skill-root` 保持兼容；`all` 先检查所有内容冲突 | REG-0058 | 禁用新路由参数即可退回 CR-0015 的 Codex 默认目录；不删除或覆盖已安装 Skill | 定向 46 passed、Skill validator、Project Architect strict 与隔离 wheel 安装通过；BUG-0023 已修复固定片段编译与 SHA 回归；全量 358 passed/3 skipped；发布追溯 REL-0013 |
 | CR-0028 | 补齐 SG JSON 创建后的 ASE 图整理工作流：端口级复用计划、折叠接口计算基线、刚性计算岛排布与分层验收 | MOD-CORE MOD-LAYOUT MOD-CLI MOD-SKILL | 只读 graph-review；fanout 必须显式选择，默认消费组规则不变；island-columns 默认 1；不重建现有 Shader，不修改 SGCLI，不猜动态节点字段 | REG-0060 REG-0061 | 不用新增参数即保留原布局；回退本任务文件增量，不覆盖并行创建器改动；不发布 | 自动验证中；Local Var 自动 Editor 创建未实现，画布/GUI/渲染验收待具体目标，不以只读计划冒充写入 |
 | CR-0029 | 强化 CLI JSON 输出契约：原子 ASCII-safe 写出、拒绝非有限数、统一契约/命令/版本字段与错误码注册表 | MOD-CORE MOD-LAYOUT MOD-BRIDGE MOD-CLI MOD-SKILL | 除显式 `--help` 外 stdout 恒为一个严格 JSON；`--version` 改为 JSON；新增顶层字段为加性；损坏几何失败关闭 | REG-0062 | 回滚到 `v0.6.4`；不改 Shader 计算、Editor 交易或用户资产 | 对抗审计 5 项 finding 均关闭；定向 76 passed，全量 388 passed/3 skipped，wheel/ASCII/中文错误路径 smoke 通过；发布见 REL-0015 |
-| CR-0030 | 两 CLI 取消外部中间包：ASECLI 直接输出 SGCLI 原生裸 `sgcli.native.v3`，并以方向化 spec/report 文件名与 SGCLI 的 EditorGraphSpec 输出对称 | MOD-CLI MOD-BRIDGE MOD-SG-EXPORT MOD-CONTRACT MOD-SKILL | 新增 `export-sg`、`contract`；`create --spec` 拒绝 wrapper/report/反方向后缀；旧历史 JSON 不进入正式流程；候选 wheel Schema/篡改负例、发布 receipt 与跨仓 required workflow 失败关闭；不调用或依赖 SGCLI 运行时 | REG-0063 | 禁用新入口或回滚 wheel；不得恢复消费者隐式解包/正式 translator，不批量改名历史文件 | `v0.8.0` 发布候选：第三轮补齐只读 Deploy Key、精确 SGCLI commit、build-once wheel、WaveNoise 语义 fixture、receipt 交叉绑定、fixture-set 完整性、Draft 数值等价与 clean provenance；全量 `428 passed, 3 skipped`（Python 3.10 同结果）。远端发布与本机安装执行中；最终 wheel Editor/画布/渲染仍开放 |
+| CR-0030 | 两 CLI 取消外部中间包：ASECLI 直接输出 SGCLI 原生裸 `sgcli.native.v3`，并以方向化 spec/report 文件名与 SGCLI 的 EditorGraphSpec 输出对称 | MOD-CLI MOD-BRIDGE MOD-SG-EXPORT MOD-CONTRACT MOD-SKILL | 新增 `export-sg`、`contract`；`create --spec` 拒绝 wrapper/report/反方向后缀；旧历史 JSON 不进入正式流程；候选 wheel Schema/篡改负例、发布 receipt 与跨仓 required workflow 失败关闭；不调用或依赖 SGCLI 运行时 | REG-0063 | 禁用新入口或回滚 wheel；不得恢复消费者隐式解包/正式 translator，不批量改名历史文件 | `v0.8.0@58ccf10` 已发布：publish run 34592028675、GitHub/PyPI 同源资产、本机 CLI/Skill 和 WaveNoise 双向原文件 SHA 全部复核通过；全量 `428 passed, 3 skipped`（Python 3.10 同结果）。最终 wheel Editor/画布/渲染仍开放，不声明视觉等价 |
 
 授权来源：用户于 2026-09-01 明确要求“执行优化任务，全部完成”，并于同日批准吸收 ASE Editor API 方案后要求“执行”；随后明确要求先推送并执行对抗性审计优化计划。2026-09-02 用户要求推送并进一步明确“以正式的 CLI 形式发布”，据此创建私有正式 GitHub Release `v0.1.0`；同日进一步明确说明条样式一致性必须成为 CLI 的规范能力，据此登记 CR-0011/BUG-0015，并明确所有通过 CLI 构建的 ASE 文件都必须使用中文属性显示名、自动变量名与默认值 Tooltip 和中文使用说明，据此登记 CR-0012。同日用户要求执行 `2026-09-02-211218` 优化计划，据此登记 CR-0013；随后书面授权打 tag 并发布 `v0.2.0`。2026-09-03 用户要求执行最新优化任务书并把剩余门槛全部完成，据此登记 CR-0014 并授权完成必要提交、直接推送与远程 CI 验证；随后明确要求“推送并发布”，据此授权创建私有正式版本 `v0.3.0`。2026-09-04 用户要求将 ASE 精排与 Inspector 编排规范作为可安装 Skill 并授权发布 `v0.3.1`。2026-09-05 用户要求执行 MZGUI 对接与效率优化，随后分别授权私有版本 `v0.4.0`、`v0.4.1` 和审计补丁 `v0.4.2`。2026-09-06 用户明确 Tooltip 必填、HelpBox 可选、条件置灰并授权 `v0.5.0`；随后要求脑图式 DAG 精排并授权 `v0.6.0`；又将 Local Var 门槛改为两个及以上不同消费组，据此登记 CR-0020；最后以多组人工示范明确递归局部鱼骨、Comment 保留和显式 WireNode 路由并要求执行，据此登记 CR-0021。2026-09-07 用户明确授权提交、推送、私有 Release `v0.6.1` 和本机安装。同日进一步明确要求开源该项目并提供同事安装命令，据此登记 CR-0022：采用 MIT 许可证并将 GitHub 仓库转为 public。随后要求做成最终、最常见的安装形态，据此登记 CR-0023：发布 PyPI 项目 `asecli`，用户命令为 `uv tool install asecli`；仍未授权 CLI Hub。2026-09-08 用户进一步明确要求推送、发布新版本并安装本机，据此授权 `v0.6.2` 的 main 推送、tag、GitHub Release、PyPI Trusted Publishing 与本机正式产物安装；CLI Hub 仍不在授权范围。首次 PyPI OIDC 因没有对应账号和 Trusted Publisher 失败后，用户说明没有 PyPI 账号并明确要求继续 GitHub Release 发布，据此以 CR-0025 将 PyPI 延后并完成 GitHub-only 交付。随后用户取得 PyPI 账号并明确要求继续配置，据此登记 CR-0026，启用精确匹配的 Pending Publisher 并通过 run 34192454152 attempt 3 完成首次 PyPI 发布。同日用户要求将现有 Skill 改为主流 Agent 通用，据此登记 CR-0027；本次不包含推送、tag 或发布。2026-09-10 用户要求对 JSON 输出能力执行对抗审计与优化，随后明确要求推送、发布并安装本机，据此登记 CR-0029 并授权 `v0.7.0` GitHub Release、PyPI Trusted Publishing 与本机 CLI/Skill 安装；CLI Hub 仍不在范围。

{asecli-0.8.0 → asecli-0.8.1}/docs/00-governance/timeline.md

@@ -95,4 +95,5 @@
 | 2026-09-10 | v0.7.0 双渠道发布与本机安装完成 | tag `v0.7.0` 解引用为 `029fea4`；run 34475832000 的双 Python verify、package 与 PyPI OIDC publish 全成功；GitHub Release 为 Latest，五项资产与 PyPI wheel/sdist 摘要一致；本机 CLI、官方索引 uvx 及三处全局 Skill 均通过，旧 Skill 已完整备份 | REL-0015 CR-0029 TASK-0064 REG-0062 |
 | 2026-09-11 | 双向裸规格直连第二轮对抗整改 | CR-0030 / ADR-0022 / TASK-0065 / REG-0063：完成 AA-OPT-201～207，补齐 Draft parity、候选 wheel 契约可信根、篡改负例、required workflow、语义矩阵、离线 artifact、发布 receipt 与迁移 provenance。全量 `419 passed, 3 skipped`；最终 wheel `2e38ef11...` 可复现，真实双向 WaveNoise producer/consumer SHA 一致。用户级 ASECLI Skill 因不同内容按规则拒绝覆盖；最终 wheel Editor/画布/render 仍开放。未提交、未推送、未发布。 | CR-0030 ADR-0022 TASK-0065 REG-0063 AA-OPT-201 AA-OPT-202 AA-OPT-203 AA-OPT-204 AA-OPT-205 AA-OPT-206 AA-OPT-207 |
 | 2026-09-11 | 双向裸规格直连第三轮对抗整改 | 执行 AA-OPT-301～308：public PR 不读取 private SGCLI；可信链要求只读 token 和精确 40 位 SHA；package wheel 原样交给 cross-CLI/publish；WaveNoise 双向语义 fixture、receipt 绑定、fixture-set 完整性、Draft 数值等价与 clean provenance 通过。全量及 Python 3.10 均 `428 passed, 3 skipped`；本地 clean release-mode 双 wheel matrix/verifier 通过。AA-OPT-309 的远端 ruleset/required checks、最终 wheel Editor/画布/render 仍开放。未提交、未推送、未发布。 | CR-0030 TASK-0065 REG-0063 AA-OPT-301 AA-OPT-302 AA-OPT-303 AA-OPT-304 AA-OPT-305 AA-OPT-306 AA-OPT-307 AA-OPT-308 AA-OPT-309 |
-| 2026-09-11 | v0.8.0 发布候选 | 用户授权两个 CLI 分别推送、发布并更新本机；ASECLI 以 pre-1.0 次版本发布双向裸 JSON 与第三轮门禁，保留 v0.7.0 回滚点。私有 SGCLI 跨仓认证使用专用只读 Deploy Key；本地双 Python、可复现构建和 clean release-mode 双 wheel matrix/verifier 已通过。远端 publish/OIDC、双渠道回读与本机安装继续执行；最终 Editor/画布/render 保持开放披露。 | REL-0016 CR-0030 TASK-0065 REG-0063 |
+| 2026-09-11 | v0.8.0 双渠道发布与本机安装完成 | tag `v0.8.0` 解引用到 `58ccf10`；[publish run 34592028675](https://github.com/seeseeczl/ASECLI/actions/runs/34592028675) 的双 Python、package、public contract、Deploy Key cross-CLI 和 PyPI OIDC 全绿。GitHub Release 五项资产与 PyPI wheel/sdist 摘要一致且未 yanked；本机 CLI 为 0.8.0，`.agents`/`.claude` Skill 与 wheel 一致，旧三处 Skill 已备份。WaveNoise 双向生产/消费 SHA 一致；最终 wheel 的 Editor/画布/render 保持未验证披露。 | REL-0016 CR-0030 TASK-0065 REG-0063 |
+| 2026-09-14 | v0.8.1 发布候选 | matcher-v4 识别并折叠原生 `Lerp(white,color,alpha)`，保留 Wave Noise 不可表示 Alpha Blend 的失败关闭；SGCLI 普通 authoring 与转换认证分离。发布前全量 `461 passed, 3 skipped`，CI governance、供应链与 diff 通过；远端 publish/PyPI、正式资产回读和本机安装待执行，Editor/画布/render 保持开放披露。 | REL-0017 CR-0030 TASK-0066 REG-0063 |

{asecli-0.8.0 → asecli-0.8.1}/docs/00-governance/traceability.csv

@@ -62,4 +62,4 @@ CR-0027,变更,已发布,ADR-0021,MOD-CLI MOD-SKILL,TASK-0059 TASK-0060,REG-0058
 BUG-0023,缺陷,已修复,ADR-0020,MOD-BRIDGE,TASK-0060,REG-0059,2026-09-08 两类真实 v2 CLI create/manifest/validate/recompile 通过；358 passed/3 skipped,REL-0013,long,方法体兼容 KeyValuePair；SphereMask Radius 重载仍为已知限制
 CR-0028,变更,已发布（视觉与自动LocalVar仍待验）,ADR-0021,MOD-CORE MOD-LAYOUT MOD-CLI MOD-SKILL,TASK-0061 TASK-0062 TASK-0063,REG-0060 REG-0061,端口级基线与复用计划及岛内刚性多列布局；未自动创建LocalVar；未做实机验收,REL-0014,long,保留并行创建器改动；不修改SGCLI
 CR-0029,变更,已发布,ADR-0001 ADR-0009,MOD-CORE MOD-LAYOUT MOD-BRIDGE MOD-CLI MOD-SKILL,TASK-0064,REG-0062,2026-09-10 v0.7.0@029fea4；run 34475832000 全成功；GitHub/PyPI 摘要一致；本机 CLI/Skill 安装幂等,REL-0015,long,严格原子 JSON；非有限数拒绝；version JSON；22 错误码；17 子命令动态覆盖
-CR-0030,变更,发布候选（视觉门禁开放）,ADR-0022,MOD-CLI MOD-BRIDGE MOD-SG-EXPORT MOD-CONTRACT MOD-SKILL,TASK-0065,REG-0063,2026-09-11 v0.8.0 候选 428 passed/3 skipped；Python 3.10 同结果；clean release-mode 双 wheel WaveNoise/receipt/Schema/artifact verifier 通过,REL-0016,long,消费者原生裸规格与方向命名；远端发布与本机安装执行中
+CR-0030,变更,已发布（视觉门禁开放）,ADR-0022,MOD-CLI MOD-BRIDGE MOD-SG-EXPORT MOD-CONTRACT MOD-SKILL,TASK-0065,REG-0063,2026-09-11 v0.8.0 publish run 34592028675 全绿；GitHub/PyPI 资产、本机 CLI/Skill 和 WaveNoise 双向原文件 SHA 通过,REL-0016,long,消费者原生裸规格与方向命名；最终 wheel Editor/画布/渲染仍开放

{asecli-0.8.0 → asecli-0.8.1}/docs/04-delivery/project-plan-task-charter.md

@@ -105,7 +105,8 @@ kickoff_completion: complete
 | TASK-0062 | CR-0028 | 独立计算岛刚性多列摆放 | TASK-0061；已有鱼骨与真实几何契约 | island_layout.py/meticulous_layout.py 与测试 | 拆改岛内连接、自动 WireNode 路由、虚构几何 | long/2026-09-08 | REG-0061；相对位置、依赖与二次幂等 | 几何缺失/布局硬失败时零写入 | 定向 94 passed；自动环节已验证；实机待验 | 自动验证完成，实机待验 |
 | TASK-0063 | CR-0028 | 将转换后交付流程接入 CLI 与 Skill | TASK-0061 TASK-0062；用户完整过程 | graph_review_command.py/main.py/layout_command.py；Skill/README/治理 | 发布、覆盖用户 Skill、改并行创建器、生产 Shader | long/2026-09-08 | REG-0060 REG-0061；CLI JSON/只读/失败码、Skill validator | 计划冒充创建或结构冒充视觉时停止 | 定向 94 passed；自动环节已验证；未接入自动 Local Var Editor 创建 | 自动验证完成，实机待验 |
 | TASK-0064 | CR-0029 | 关闭 JSON 输出对抗审计并发布安装 `v0.7.0` | AA-OPT-001～005；用户推送、发布、安装授权 | contract/main；Commentary/layout/Editor 几何；README/Skill；版本与 REL-0015 | CLI Hub；修改用户 Shader；覆盖未知 Skill；绕过发布门禁 | long/2026-09-10 | REG-0062；双 Python、完整治理、可复现构建、远程 publish/OIDC、双渠道摘要、本机 CLI/Skill | CI、OIDC、摘要、安装或 Skill 冲突任一失败即停止，保留 `v0.6.4` | v0.7.0@029fea4；run 34475832000 全成功；GitHub/PyPI 摘要一致；本机 CLI/Skill 安装与幂等通过 | 已发布 |
-| TASK-0065 | CR-0030 ADR-0022 AA-OPT-201～207 AA-OPT-301～309 REL-0016 | 实现 ASECLI→native v3 与 SGCLI→EditorGraphSpec v3 的双向裸规格原文件直连并发布 0.8.0 | 用户确认的合并优化计划、三轮对抗审计与推送发布安装授权 | `export_sg.py`、`sg_export/**`、`contracts/**`、`schema_validation.py`、`conversion_output.py`、CLI/create、README/Skill、测试、治理和 cross-CLI workflows | 调用 SGCLI；恢复 source-native/wrapper 主链；修改生产 Shader；覆盖用户 dirty Skill；绕过失败门禁 | long/2026-09-11 | REG-0063；Draft parity；候选 wheel Schema/篡改负例；固定 WaveNoise 语义；artifact 完整性与 clean provenance；双 Python全量；远端 publish/OIDC；双渠道回读；本机 CLI/Skill；Editor/画布/渲染分层 | 未知 ASE/Target/资源/端口、部分 spec、同名覆盖、Schema 漂移、脏候选、远端门禁或资产冲突即停止 | 0.8.0 候选全量 428 passed/3 skipped；0.3.0/0.8.0 wheel 可复现且本地 clean matrix/verifier 通过；只读 Deploy Key 已配置；远端发布与安装执行中；最终 Editor/画布/渲染开放披露 | 发布候选（REL-0016） |
+| TASK-0065 | CR-0030 ADR-0022 AA-OPT-201～207 AA-OPT-301～309 REL-0016 | 实现 ASECLI→native v3 与 SGCLI→EditorGraphSpec v3 的双向裸规格原文件直连并发布 0.8.0 | 用户确认的合并优化计划、三轮对抗审计与推送发布安装授权 | `export_sg.py`、`sg_export/**`、`contracts/**`、`schema_validation.py`、`conversion_output.py`、CLI/create、README/Skill、测试、治理和 cross-CLI workflows | 调用 SGCLI；恢复 source-native/wrapper 主链；修改生产 Shader；覆盖用户 dirty Skill；绕过失败门禁 | long/2026-09-11 | REG-0063；Draft parity；候选 wheel Schema/篡改负例；固定 WaveNoise 语义；artifact 完整性与 clean provenance；双 Python全量；远端 publish/OIDC；双渠道回读；本机 CLI/Skill；Editor/画布/渲染分层 | 未知 ASE/Target/资源/端口、部分 spec、同名覆盖、Schema 漂移、脏候选、远端门禁或资产冲突即停止 | `v0.8.0@58ccf10` 已发布；run 34592028675、GitHub/PyPI 资产、本机 CLI/Skill 和 WaveNoise 双向原文件 SHA 全部复核通过；最终 Editor/画布/渲染开放披露 | 已发布（REL-0016，视觉开放） |
+| TASK-0066 | CR-0030 ADR-0022 AA-OPT-201 AA-OPT-202 REL-0017 | 发布并安装 matcher-v4 与 ASE→SG 转换认证补丁 v0.8.1 | 用户要求两个 CLI 分别推送、发布并安装本机 | `sg_export/**`、report、版本、README、回归与发布证据 | 修改生产 Shader；绕过不可表示 Alpha Blend；恢复 wrapper；覆盖未知 Skill | long/2026-09-14 | REG-0063；两仓全量、治理、供应链、Tag publish、GitHub/PyPI 摘要、本机 CLI/Skill；Editor/视觉分层 | 任一自动/远端/摘要/安装门禁失败即停止并保留 v0.8.0 | 本地候选 `461 passed, 3 skipped`，远端与安装待执行 | verified |
 
 ## 验证、风险与回滚
 

{asecli-0.8.0 → asecli-0.8.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "asecli"
-version = "0.8.0"
+version = "0.8.1"
 description = "Agent-native CLI for reading, modifying, validating and compiling Amplify Shader Editor assets"
 readme = "README.md"
 requires-python = ">=3.10"

{asecli-0.8.0 → asecli-0.8.1}/scripts/install.ps1

@@ -21,7 +21,7 @@ if ($env:ASECLI_REF) {
     $tag = (Invoke-RestMethod "https://api.github.com/repos/$repo/releases/latest").tag_name
 }
 if (-not $tag) {
-    $tag = "v0.8.0"
+    $tag = "v0.8.1"
 }
 
 $version = $tag.TrimStart("v")

{asecli-0.8.0 → asecli-0.8.1}/scripts/install.sh

@@ -32,7 +32,7 @@ else
   TAG=$(curl -fsSL "https://api.github.com/repos/${REPO}/releases/latest" | sed -n 's/.*"tag_name": *"\([^"]*\)".*/\1/p' | head -n 1)
 fi
 if [ -z "$TAG" ]; then
-  TAG="v0.8.0"
+  TAG="v0.8.1"
 fi
 VERSION=${TAG#v}
 WHEEL="https://github.com/${REPO}/releases/download/${TAG}/asecli-${VERSION}-py3-none-any.whl"

{asecli-0.8.0 → asecli-0.8.1}/src/asecli/__init__.py

@@ -1,3 +1,3 @@
 """AseCLI — agent-native toolkit for Amplify Shader Editor assets."""
 
-__version__ = "0.8.0"
+__version__ = "0.8.1"

asecli-0.8.1/src/asecli/sg_export/alpha_modulate.py

@@ -0,0 +1,113 @@
+"""Recognize the certified single-assignment AlphaModulate HLSL subset."""
+
+from __future__ import annotations
+
+import re
+
+
+def inputs(function: dict | None) -> tuple[str, str] | None:
+    if not function or function.get("source") != "String":
+        return None
+    ports = function.get("ports", [])
+    outputs = [port for port in ports if port.get("direction") == "Output"]
+    if len(outputs) != 1 or outputs[0].get("type") != "Vector3":
+        return None
+    output_name = outputs[0].get("name")
+    if not isinstance(output_name, str) or not re.fullmatch(r"[A-Za-z_]\w*", output_name):
+        return None
+    body = _strip_comments(function.get("body", ""))
+    if body is None:
+        return None
+    match = re.fullmatch(
+        rf"\s*{re.escape(output_name)}\s*=\s*lerp\s*\((.*)\)\s*;\s*",
+        body,
+        flags=re.DOTALL,
+    )
+    if not match:
+        return None
+    args = _split_arguments(match.group(1))
+    if len(args) != 3 or not _is_white(args[0]):
+        return None
+    color_name, alpha_name = args[1].strip(), args[2].strip()
+    if not re.fullmatch(r"[A-Za-z_]\w*", color_name):
+        return None
+    if not re.fullmatch(r"[A-Za-z_]\w*", alpha_name):
+        return None
+    input_types = {
+        port.get("name"): port.get("type")
+        for port in ports if port.get("direction") == "Input"
+    }
+    if input_types.get(color_name) != "Vector3" or input_types.get(alpha_name) != "Vector1":
+        return None
+    return color_name, alpha_name
+
+
+def _split_arguments(value: str) -> list[str]:
+    result, start, depth = [], 0, 0
+    for index, char in enumerate(value):
+        if char == "(":
+            depth += 1
+        elif char == ")":
+            depth -= 1
+            if depth < 0:
+                return []
+        elif char == "," and depth == 0:
+            result.append(value[start:index])
+            start = index + 1
+    if depth != 0:
+        return []
+    result.append(value[start:])
+    return result
+
+
+def _is_white(value: str) -> bool:
+    compact = re.sub(r"\s+", "", value)
+    constructor = re.fullmatch(r"(?:float|half)3\((.*)\)", compact, flags=re.IGNORECASE)
+    if constructor:
+        parts = _split_arguments(constructor.group(1))
+        return len(parts) == 3 and all(_is_one(part) for part in parts)
+    scalar = re.fullmatch(r"\(?([^()]+)\)?\.xxx", compact, flags=re.IGNORECASE)
+    return bool(scalar and _is_one(scalar.group(1)))
+
+
+def _is_one(value: str) -> bool:
+    try:
+        return float(value.rstrip("fF")) == 1.0
+    except ValueError:
+        return False
+
+
+def _strip_comments(value: str) -> str | None:
+    result, index, state = [], 0, "code"
+    while index < len(value):
+        char = value[index]
+        following = value[index + 1] if index + 1 < len(value) else ""
+        if state == "code":
+            if char == '"':
+                state = "string"
+                result.append(char)
+            elif char == "/" and following in {"/", "*"}:
+                result.extend("  ")
+                index += 1
+                state = "line_comment" if following == "/" else "block_comment"
+            else:
+                result.append(char)
+        elif state == "string":
+            result.append(char)
+            if char == "\\" and following:
+                result.append(following)
+                index += 1
+            elif char == '"':
+                state = "code"
+        elif state == "line_comment":
+            result.append(char if char in "\r\n" else " ")
+            if char in "\r\n":
+                state = "code"
+        elif char == "*" and following == "/":
+            result.extend("  ")
+            index += 1
+            state = "code"
+        else:
+            result.append(char if char in "\r\n" else " ")
+        index += 1
+    return None if state in {"string", "block_comment"} else "".join(result)

{asecli-0.8.0 → asecli-0.8.1}/src/asecli/sg_export/conversion_report.py

@@ -7,6 +7,8 @@ import json
 from pathlib import Path
 from typing import Any
 
+from .report_semantics import check_evidence
+
 
 CHECK_RULES = {
     "properties": "SEM-PROP-001",
@@ -74,7 +76,7 @@ def build_public_report(
         name: {
             "rule": rule,
             "status": check_status[name],
-            "evidence": _check_evidence(name, check_status[name], internal, failure),
+            "evidence": check_evidence(name, check_status[name], internal, failure),
         }
         for name, rule in CHECK_RULES.items()
     }
@@ -188,6 +190,9 @@ def _alpha_equation_verified(internal: dict[str, Any] | None) -> bool:
     target = internal.get("target_mapping", {})
     if target.get("surface") == "Opaque":
         return True
+    semantics = internal.get("surface_semantics", {})
+    if semantics.get("alpha_blend_equivalent") is not True or semantics.get("unresolved"):
+        return False
     return any(
         isinstance(item, dict)
         and item.get("block") == "SurfaceDescription.Alpha"
@@ -199,31 +204,16 @@ def _blend_equation_verified(internal: dict[str, Any] | None) -> bool:
     if not isinstance(internal, dict):
         return False
     target = internal.get("target_mapping", {})
+    semantics = internal.get("surface_semantics", {})
+    if target.get("surface") == "Transparent" and (
+        semantics.get("rgb_blend_equivalent") is not True
+        or semantics.get("alpha_blend_equivalent") is not True
+    ):
+        return False
+    if semantics.get("unresolved"):
+        return False
+    if semantics.get("target_pipeline", {}).get("implicit_alpha_modulate"):
+        source_explicit = semantics.get("source", {}).get("explicit_alpha_modulate", False)
+        if source_explicit and not semantics.get("rewrites"):
+            return False
     return target.get("blend") in {"Alpha", "Premultiply", "Additive", "Multiply"}
-
-
-def _check_evidence(
-    name: str,
-    state: str,
-    internal: dict[str, Any] | None,
-    failure: str | None,
-) -> str:
-    if state == "verified":
-        return {
-            "properties": "ASE properties reconciled with ShaderLab declarations",
-            "resources": "all texture dependencies are null or have matching importer evidence",
-            "graph_algorithm": "certified ASE nodes, defaults and wires were deterministically mapped",
-            "target": "certified URP Unlit Master settings were mapped",
-            "passes": "the certified ASE URP Unlit pass set was recognized",
-            "alpha_equation": "opaque alpha behavior or the explicit Alpha block binding was preserved",
-            "blend_equation": "the certified ASE Master blend mode was mapped to the SG target",
-        }[name]
-    if failure:
-        return f"conversion stopped before this invariant was proved: {failure}"
-    diagnostics = internal.get("diagnostics", []) if isinstance(internal, dict) else []
-    if diagnostics:
-        first = diagnostics[0]
-        return f"{first.get('code', 'UNPROVEN')}: {first.get('reason', 'not proven')}"
-    if name in {"alpha_equation", "blend_equation"}:
-        return "the source target equation could not be proved from the certified Master mapping"
-    return "not proven by the producer-only export"

{asecli-0.8.0 → asecli-0.8.1}/src/asecli/sg_export/decoder.py

@@ -14,6 +14,7 @@ from .model import (
 from .nodes import convert_node
 from .sources import compiled_properties, load_asset_map
 from .evidence import inspector_degradations, record_port, validate_template_passes
+from .surface_semantics import normalize_surface_outputs
 from .reconciliation import (
     groups, reconcile_shaderlab_properties, unique_properties, verify_texture_dependencies,
 )
@@ -123,6 +124,7 @@ def build_candidate(
     for node in nodes:
         for target_id in node.source_ids:
             edges.append(SemanticEdge(node.target_id, 0, target_id, -1))
+    nodes, edges = normalize_surface_outputs(nodes, edges, target_spec, mappings, report, text)
     groups_spec = groups(ase, mappings, diagnostics)
     report["output_bindings"] = [
         {"block": edge.target_id, "port": edge.target_port}
@@ -243,8 +245,6 @@ def _template_option(fields, label, choices, diagnostics, node_id):
 
 
 def _template_bool(fields, label, diagnostics, node_id):
-    value = _template_option(fields, label, (False, True), diagnostics, node_id)
-    return value
-
+    return _template_option(fields, label, (False, True), diagnostics, node_id)
 def _graph_precision(master):
     return "Half" if master and len(master.raw_fields) > 4 and master.raw_fields[4] == "Half" else "Single"

{asecli-0.8.0 → asecli-0.8.1}/src/asecli/sg_export/model.py

@@ -12,7 +12,7 @@ from typing import Any
 from .. import __version__
 
 REPORT_SCHEMA = "sgcli.shader-conversion-report.v1"
-MAPPING_VERSION = 3
+MAPPING_VERSION = 4
 SUPPORTED_ASE_VERSIONS = frozenset({"19109", "19602"})
 MASTER_TYPE = "AmplifyShaderEditor.TemplateMultiPassMasterNode"
 WIRE_TYPE = "AmplifyShaderEditor.WireNode"

asecli-0.8.1/src/asecli/sg_export/pass_blend.py

@@ -0,0 +1,200 @@
+"""ShaderLab and standard URP Shader Graph blend-state semantics."""
+
+from __future__ import annotations
+
+import re
+
+
+_BLEND_FACTORS = {
+    value.lower(): value
+    for value in (
+        "Zero", "One", "SrcColor", "OneMinusSrcColor", "SrcAlpha",
+        "OneMinusSrcAlpha", "DstColor", "OneMinusDstColor", "DstAlpha",
+        "OneMinusDstAlpha", "SrcAlphaSaturate",
+    )
+}
+
+
+def record_pass_blend(semantics: dict, target: dict, source_text: str | None) -> None:
+    evidence = forward_blend_evidence(source_text or "")
+    source = evidence.get("blend")
+    target_blend = target_blend_factors(target)
+    semantics["source_pass_blend"] = source
+    semantics["target_pass_blend"] = target_blend
+    semantics["pass_blend_parser"] = {
+        key: value for key, value in evidence.items() if key != "blend"
+    }
+    if source is not None and target_blend is not None:
+        semantics["rgb_blend_equivalent"] = source["rgb"] == target_blend["rgb"]
+        semantics["alpha_blend_equivalent"] = source["alpha"] == target_blend["alpha"]
+
+
+def forward_blend(source_text: str) -> dict | None:
+    """Return the unique Forward pass blend factors, or None when unproved."""
+    return forward_blend_evidence(source_text).get("blend")
+
+
+def forward_blend_evidence(source_text: str) -> dict:
+    """Parse a unique Forward Pass without trusting comments or unrelated text."""
+    masked, error = _mask_comments(source_text)
+    if error:
+        return _evidence("malformed", 0, error)
+    bodies, error = _pass_bodies(masked)
+    if error:
+        return _evidence("malformed", 0, error)
+    search_bodies = bodies or [masked]
+    forward = [body for body in search_bodies if _is_forward(body)]
+    if not forward:
+        return _evidence("missing", 0, "no Forward pass was found")
+    if len(forward) != 1:
+        return _evidence(
+            "ambiguous", len(forward), f"expected one Forward pass, found {len(forward)}"
+        )
+    blend, status, reason, directive_count = _blend_in_pass(forward[0])
+    result = _evidence(status, 1, reason)
+    result["directive_count"] = directive_count
+    if blend is not None:
+        result["blend"] = blend
+    return result
+
+
+def target_blend_factors(target: dict) -> dict | None:
+    if target.get("surface") != "Transparent":
+        return {"rgb": ["One", "Zero"], "alpha": ["One", "Zero"]}
+    return {
+        "Alpha": {"rgb": ["SrcAlpha", "OneMinusSrcAlpha"], "alpha": ["One", "OneMinusSrcAlpha"]},
+        "Premultiply": {"rgb": ["One", "OneMinusSrcAlpha"], "alpha": ["One", "OneMinusSrcAlpha"]},
+        "Additive": {"rgb": ["SrcAlpha", "One"], "alpha": ["One", "One"]},
+        "Multiply": {"rgb": ["DstColor", "Zero"], "alpha": ["Zero", "One"]},
+    }.get(target.get("blend"))
+
+
+def _evidence(status: str, count: int, reason: str) -> dict:
+    return {
+        "status": status,
+        "candidate_count": count,
+        "reason": reason,
+        "parser_version": 2,
+    }
+
+
+def _mask_comments(source: str) -> tuple[str, str | None]:
+    """Replace comments with spaces while preserving offsets, strings and newlines."""
+    chars = list(source)
+    index, state = 0, "code"
+    while index < len(source):
+        char = source[index]
+        following = source[index + 1] if index + 1 < len(source) else ""
+        if state == "code":
+            if char == '"':
+                state = "string"
+            elif char == "/" and following == "/":
+                chars[index] = chars[index + 1] = " "
+                index += 1
+                state = "line_comment"
+            elif char == "/" and following == "*":
+                chars[index] = chars[index + 1] = " "
+                index += 1
+                state = "block_comment"
+        elif state == "string":
+            if char == "\\" and following:
+                index += 1
+            elif char == '"':
+                state = "code"
+        elif state == "line_comment":
+            if char in "\r\n":
+                state = "code"
+            else:
+                chars[index] = " "
+        else:
+            if char == "*" and following == "/":
+                chars[index] = chars[index + 1] = " "
+                index += 1
+                state = "code"
+            elif char not in "\r\n":
+                chars[index] = " "
+        index += 1
+    if state == "block_comment":
+        return "".join(chars), "unterminated block comment"
+    if state == "string":
+        return "".join(chars), "unterminated string literal"
+    return "".join(chars), None
+
+
+def _mask_strings(source: str) -> str:
+    chars = list(source)
+    index, quoted = 0, False
+    while index < len(source):
+        char = source[index]
+        if not quoted and char == '"':
+            chars[index] = " "
+            quoted = True
+        elif quoted:
+            if char == "\\" and index + 1 < len(source):
+                chars[index] = chars[index + 1] = " "
+                index += 1
+            else:
+                if char == '"':
+                    quoted = False
+                if char not in "\r\n":
+                    chars[index] = " "
+        index += 1
+    return "".join(chars)
+
+
+def _pass_bodies(source: str) -> tuple[list[str], str | None]:
+    structural = _mask_strings(source)
+    result, covered_until = [], -1
+    for match in re.finditer(r"\bPass\s*\{", structural, flags=re.IGNORECASE):
+        if match.start() < covered_until:
+            continue
+        opening = structural.find("{", match.start(), match.end())
+        depth = 0
+        for index in range(opening, len(structural)):
+            if structural[index] == "{":
+                depth += 1
+            elif structural[index] == "}":
+                depth -= 1
+                if depth == 0:
+                    result.append(source[opening + 1:index])
+                    covered_until = index + 1
+                    break
+        else:
+            return [], "unterminated Pass block"
+    return result, None
+
+
+def _is_forward(body: str) -> bool:
+    header = re.split(r"\b(?:HLSLPROGRAM|CGPROGRAM)\b", body, maxsplit=1, flags=re.IGNORECASE)[0]
+    return bool(re.search(r'\bName\s+"Forward"(?!\w)', header, flags=re.IGNORECASE))
+
+
+def _blend_in_pass(body: str) -> tuple[dict | None, str, str, int]:
+    header = re.split(r"\b(?:HLSLPROGRAM|CGPROGRAM)\b", body, maxsplit=1, flags=re.IGNORECASE)[0]
+    directives = re.findall(r"(?im)^\s*Blend\s+(.+?)\s*$", header)
+    if len(directives) > 1:
+        return None, "ambiguous", f"expected at most one Blend directive, found {len(directives)}", len(directives)
+    if not directives:
+        if re.search(r"\bBlend\b", header, flags=re.IGNORECASE):
+            return None, "unsupported", "Blend directive is not on a uniquely parseable line", 1
+        return _off_blend(), "parsed", "Forward pass uses implicit Blend Off", 0
+    value = directives[0].strip()
+    if value.lower() == "off":
+        return _off_blend(), "parsed", "Forward pass explicitly uses Blend Off", 1
+    channels = [part.strip().split() for part in value.split(",")]
+    if len(channels) not in (1, 2) or any(len(part) != 2 for part in channels):
+        return None, "unsupported", f"unsupported Blend directive: {value}", 1
+    normalized = []
+    for source, destination in channels:
+        pair = (_BLEND_FACTORS.get(source.lower()), _BLEND_FACTORS.get(destination.lower()))
+        if None in pair:
+            return None, "unsupported", f"unsupported Blend factors: {value}", 1
+        normalized.append(list(pair))
+    return {
+        "rgb": normalized[0],
+        "alpha": normalized[1] if len(normalized) == 2 else list(normalized[0]),
+    }, "parsed", "unique Forward pass Blend parsed", 1
+
+
+def _off_blend() -> dict:
+    return {"rgb": ["One", "Zero"], "alpha": ["One", "Zero"]}