docs: polish v0.1.2 public collection guidance

CATALOG.zh-CN.md

@@ -10,11 +10,11 @@
 | --- | --- |
 | 集合 | Personal Skills Collection |
 | 包数量 | 5 个第一方 Skill |
-| 当前状态 | v0.1.1 已发布；独立 `review-loop agent-skill` acceptance 仍为 BLOCKED |
+| 当前状态 | v0.1.1 已发布且可安装；独立 `review-loop agent-skill` 证据仍为 `BLOCKED` |
 | 稳定版本 | [v0.1.1](https://github.com/LightDevCoder/skills/releases/tag/v0.1.1) |
 | 安装权威 | [docs/INSTALLATION.zh-CN.md](docs/INSTALLATION.zh-CN.md) |
 | 发现检查 | [collection-discovery-tests.ps1](tests/collection-discovery-tests.ps1) |
-| 证据 | [v0.1.1 release evidence](docs/evidence/releases/v0.1.1/) |
+| 证据 | [v0.1.1 发布证据](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.zh-CN.md) |
 
 ## 已准入 Skill
 
@@ -30,7 +30,7 @@
 ### project-init
 
 - **作用：** 从最小 preset 初始化项目，保留既有指令并验证结果路径。
-- **调用：** User-invoked only。
+- **调用：** 仅 user-invoked。
 - **包：** [skills/project-init/](skills/project-init/)
 - **状态：** 第一方已准入。
 - **证据：** [package tests](skills/project-init/tests/)；[使用指南](docs/zh-CN/skills/project-init.md)。
@@ -39,7 +39,7 @@
 ### ask-light
 
 - **作用：** 从当前上下文推荐一个下一 Skill 或一个 bounded workflow recipe。
-- **调用：** User-invoked only；永不执行推荐结果。
+- **调用：** 仅 user-invoked；永不执行推荐结果。
 - **包：** [skills/ask-light/](skills/ask-light/)
 - **状态：** 第一方已准入；支持显式 `next` 和 `workflow` 模式。
 - **证据：** [scanner tests](skills/ask-light/tests/)；[使用指南](docs/zh-CN/skills/ask-light.md)。
@@ -48,7 +48,7 @@
 ### learn-anything
 
 - **作用：** 将证据充分的对话、笔记、workflow 或资料转成可复用 Agent Skill 方法。
-- **调用：** User-invoked only。
+- **调用：** 仅 user-invoked。
 - **包：** [skills/learn-anything/](skills/learn-anything/)
 - **状态：** 第一方已准入；保留 source-sufficiency 与 deterministic package-build 边界。
 - **证据：** [package contract](skills/learn-anything/SKILL.md)；[使用指南](docs/zh-CN/skills/learn-anything.md)。

CHANGELOG.md

@@ -38,8 +38,11 @@ drafted.
   [INSTALLATION_VERIFICATION.md](docs/evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.md).
 - Structural and package evidence:
   [TEST_SUMMARY.md](docs/evidence/releases/v0.1.1/TEST_SUMMARY.md).
-- Independent `review-loop agent-skill` acceptance remains `BLOCKED`; see
-  [release receipt](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.md).
+- The five-package collection remains installable and its collection-quality
+  checks passed. Independent evaluator evidence for the `review-loop
+  agent-skill` acceptance gate remains `BLOCKED`; this does not block ordinary
+  installation or use. See the [release receipt](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.md)
+  for the exact evidence boundary.
 
 ## 0.1.0 — 2026-07-23
 

CHANGELOG.zh-CN.md

@@ -24,9 +24,11 @@
 - 合并后的 release commit 上 GitHub Actions `collection-quality`：`PASS`。
 - 整仓和单 Skill fresh install：使用 CLI `1.5.20`，均为 `PASS`。
 - GitHub release：https://github.com/LightDevCoder/skills/releases/tag/v0.1.1
-- 整仓与单 Skill fresh-install 目标记录：[INSTALLATION_VERIFICATION.md](docs/evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.md)。
-- 结构与包测试证据：[TEST_SUMMARY.md](docs/evidence/releases/v0.1.1/TEST_SUMMARY.md)。
-- 独立 `review-loop agent-skill` acceptance 仍为 `BLOCKED`；见[发布收据](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.md)。
+- 整仓与单 Skill fresh-install 目标记录：[INSTALLATION_VERIFICATION.zh-CN.md](docs/evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.zh-CN.md)。
+- 结构与包测试证据：[TEST_SUMMARY.zh-CN.md](docs/evidence/releases/v0.1.1/TEST_SUMMARY.zh-CN.md)。
+- 五个包组成的集合仍可安装，`collection-quality` 检查已通过。`review-loop
+  agent-skill` acceptance gate 的独立 evaluator 证据仍为 `BLOCKED`；这不影响
+  一般安装或使用。准确证据边界见[发布收据](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.zh-CN.md)。
 
 ## 0.1.0 — 2026-07-23
 

README.md

@@ -14,8 +14,13 @@ invocation boundary, and small enough to inspect before use.
 > is published from commit `c50f1ef`. The release record and fresh-install evidence
 > live in [docs/evidence/releases/v0.1.1/](docs/evidence/releases/v0.1.1/).
 >
-> Independent `review-loop agent-skill` acceptance remains `BLOCKED`; see the
-> receipt for the exact boundary.
+> **Release status:** The collection is published and usable. Installation and
+> `collection-quality` checks passed for all five first-party Skills. The
+> remaining release-acceptance limitation is the independent evaluator evidence
+> for the `review-loop agent-skill` gate: that evidence is `BLOCKED` because a
+> fresh independent evaluator record is not attached. This does not block
+> installation or ordinary use; see the [release receipt](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.md)
+> and [limitations](docs/evidence/releases/v0.1.1/LIMITATIONS.md) for the detailed boundary.
 
 ## Quick Start
 
@@ -57,6 +62,31 @@ result. Read [Quick Start](examples/quick-start/README.md), the [Skill user
 guides](docs/skills/), and [workflow recipes](docs/workflows/) for inputs,
 outputs, handoffs, and stopping boundaries.
 
+## External capabilities
+
+The five packages above are the core collection: installing this repository
+provides all five first-party Skills listed in the catalog.
+
+Optional workflow capabilities are external or third-party dependencies and are
+not included by default:
+
+- `grill-me` / `grilling` — `grill-me` is the user-facing entry point for a
+  one-session clarification interview; it starts the underlying model-invoked
+  `grilling` capability. Treat them as one capability, not two workflow steps.
+- `research` — investigate an external fact or practice when a local preset is
+  insufficient.
+- `to-spec` — turn an approved goal and constraints into a traceable
+  specification.
+- `to-tickets` — turn an approved specification into dependency-ordered
+  tickets.
+- `implement` — carry out one bounded, unblocked implementation ticket.
+- `code-review` — provide specialist findings for a fixed change.
+- `handoff` — preserve an accepted result or blocker for closeout or resumption.
+
+These capabilities may come from `mattpocock/skills` or another external
+source. This repository does not copy them or install them automatically;
+check their availability before selecting a workflow that names them.
+
 ## First-party catalog
 
 | Skill | Purpose | Invocation | Package |

README.zh-CN.md

@@ -6,13 +6,18 @@
 
 `LightDevCoder/skills` 是公开的第一方 Agent Skills 集合，包含五个可以独立安装、独立发现、明确声明调用边界的 Skill。包内 `SKILL.md` 仍是行为权威；本 README 与用户指南负责说明使用方式。
 
-> **About：** Personal Skills Collection — Drive your creativity
+> **简介：** Personal Skills Collection — Drive your creativity
 
-> **Release：** [v0.1.1](https://github.com/LightDevCoder/skills/releases/tag/v0.1.1)
+> **发布：** [v0.1.1](https://github.com/LightDevCoder/skills/releases/tag/v0.1.1)
 > 已从 commit `c50f1ef` 发布。发布记录和 fresh-install 证据见
 > [docs/evidence/releases/v0.1.1/](docs/evidence/releases/v0.1.1/)。
 >
-> 独立 `review-loop agent-skill` acceptance 仍为 `BLOCKED`；准确边界见发布收据。
+> **发布状态：** 集合已发布且可使用。五个第一方 Skill 的安装与
+> `collection-quality` 检查均已通过。当前剩余的发布验收限制是
+> `review-loop agent-skill` 的独立 evaluator 证据：由于尚未附上 fresh
+> independent evaluator record，该项仍为 `BLOCKED`。这不影响安装或一般使用；
+> 详细边界见[发布收据](docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.zh-CN.md)
+> 和[限制说明](docs/evidence/releases/v0.1.1/LIMITATIONS.zh-CN.md)。
 
 ## Quick Start
 
@@ -28,7 +33,7 @@ npx skills add LightDevCoder/skills#v0.1.1 --yes --copy --agent codex
 npx skills add LightDevCoder/skills#v0.1.1 --skill review-loop --yes --copy --agent codex
 ```
 
-刷新 Agent host，然后在其 Skill catalog 中确认发现结果。若 host 提供文件系统，检查已安装包的 `SKILL.md` 与 `agents/openai.yaml`；脱离 source checkout 后仍能发现，才是有意义的验证。CLI 版本、destination 和结果见 [INSTALLATION_VERIFICATION.md](docs/evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.md)。
+刷新 Agent host，然后在其 Skill catalog 中确认发现结果。若 host 提供文件系统，检查已安装包的 `SKILL.md` 与 `agents/openai.yaml`；脱离 source checkout 后仍能发现，才是有意义的验证。CLI 版本、destination 和结果见 [INSTALLATION_VERIFICATION.zh-CN.md](docs/evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.zh-CN.md)。
 
 第一次不知道下一步时，推荐显式调用 `$ask-light`：
 
@@ -39,21 +44,38 @@ $ask-light next
 三个最短示例：
 
 ```text
-$ask-light next       # 推荐下一 Skill，不执行它
+$ask-light next       # 推荐下一 Skill，不执行
 $project-init         # 初始化已确认的最小项目 preset
 $review-loop init     # 冻结已有验收标准的 baseline
 ```
 
-`ask-light` 只返回建议或一个有边界的 recipe，然后停止；不会执行、安装或自动串联。阅读 [Quick Start](examples/quick-start/README.zh-CN.md)、[Skill 使用指南](docs/zh-CN/skills/)、[workflow recipes](docs/zh-CN/workflows/) 了解输入、输出、handoff 和停止点。
+`ask-light` 只返回建议或一个有边界的 recipe，然后停止；不会执行、安装或自动串联。阅读 [Quick Start](examples/quick-start/README.zh-CN.md)、[Skill 使用指南](docs/zh-CN/skills/)、[工作流 recipes](docs/zh-CN/workflows/) 了解输入、输出、handoff 和停止点。
+
+## 外部能力（External Capabilities）
+
+上面的五个包组成 Core collection：安装本仓库后即可使用目录中的五个第一方 Skill。
+
+以下是可选的 workflow 能力，来自外部或第三方来源，不属于默认集合：
+
+- `grill-me` / `grilling`：`grill-me` 是一次会话澄清访谈的用户入口，会启动底层的 model-invoked `grilling` 能力。应把它们视为同一个能力，不要当成两个重复的工作流步骤。
+- `research`：本地 preset 不足时，用于调查外部事实或实践。
+- `to-spec`：把已确认的目标和约束整理成可追踪的 specification。
+- `to-tickets`：把已批准的 specification 整理成按依赖排序的 tickets。
+- `implement`：执行一个有边界且未被阻塞的 implementation ticket。
+- `code-review`：为固定变更提供 specialist findings。
+- `handoff`：为 closeout 或后续 resume 保存已接受结果或 blocker。
+
+这些能力可能来自 `mattpocock/skills` 或其他外部来源。本仓库不复制这些 Skill，
+也不会自动安装；选择包含它们的 workflow 前，请先确认当前 host 能看到对应能力。
 
 ## 第一方目录
 
 | Skill | 作用 | 调用方式 | 包路径 |
 | --- | --- | --- | --- |
 | [review-loop](skills/review-loop/SKILL.md) | 执行有边界的证据、修复和最终验收循环。 | Model-invoked；支持手动入口。 | skills/review-loop/ |
-| [project-init](skills/project-init/SKILL.md) | 从最小 preset 初始化已确认的软件、文稿、研究、知识、数据或 Skill-development 项目。 | User-invoked only。 | skills/project-init/ |
-| [ask-light](skills/ask-light/SKILL.md) | 检查 host 并推荐一个下一 Skill 或 bounded recipe，不执行建议。 | User-invoked only。 | skills/ask-light/ |
-| [learn-anything](skills/learn-anything/SKILL.md) | 从有足够证据的资料中提炼可复用 Agent Skill 方法。 | User-invoked only。 | skills/learn-anything/ |
+| [project-init](skills/project-init/SKILL.md) | 从最小 preset 初始化已确认的软件、文稿、研究、知识、数据或 Skill-development 项目。 | 仅 user-invoked。 | skills/project-init/ |
+| [ask-light](skills/ask-light/SKILL.md) | 检查 host 并推荐一个下一 Skill 或 bounded recipe，不执行建议。 | 仅 user-invoked。 | skills/ask-light/ |
+| [learn-anything](skills/learn-anything/SKILL.md) | 从有足够证据的资料中提炼可复用 Agent Skill 方法。 | 仅 user-invoked。 | skills/learn-anything/ |
 | [manuscript-ops](skills/manuscript-ops/SKILL.md) | 治理跨格式、批次、审查和 handoff 的可复现文稿工程。 | Model-invoked；支持手动入口。 | skills/manuscript-ops/ |
 
 完整目录见 [CATALOG.zh-CN.md](CATALOG.zh-CN.md)。组合示例是文档和验证资产，不是固定 pipeline，也不是自动编排器；退休的 `project-workflow` 不会重新引入。

docs/INSTALLATION.zh-CN.md

@@ -2,7 +2,7 @@
 
 [English installation guide](INSTALLATION.md)
 
-公开第一方集合当前稳定版本是 [v0.1.1](https://github.com/LightDevCoder/skills/releases/tag/v0.1.1)，commit 为 `c50f1ef403a5f0bfe02e75d1aeff2c237556db63`。已发布 tag 的 fresh-install 证据见[release evidence](evidence/releases/v0.1.1/)。`skills/<name>/` 内的包契约仍是行为权威；本页只规定安装和验证证据。
+公开第一方集合当前稳定版本是 [v0.1.1](https://github.com/LightDevCoder/skills/releases/tag/v0.1.1)，commit 为 `c50f1ef403a5f0bfe02e75d1aeff2c237556db63`。已发布 tag 的 fresh-install 证据见[发布证据](evidence/releases/v0.1.1/RELEASE_RECEIPT.zh-CN.md)。`skills/<name>/` 内的包契约仍是行为权威；本页只规定安装和验证证据。
 
 ## Revision 语义
 
@@ -30,7 +30,7 @@ npx skills add LightDevCoder/skills#v0.1.1 --yes --copy --agent codex
 npx skills add LightDevCoder/skills#v0.1.1 --skill review-loop --yes --copy --agent codex
 ```
 
-第一条安装五个包，第二条选择同一 tag 下的完整单包。已验证的 CLI version、destination 类别、discovery 和 smoke result 见 [v0.1.1 安装记录](evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.md)。
+第一条安装五个包，第二条选择同一 tag 下的完整单包。已验证的 CLI version、destination 类别、discovery 和 smoke result 见 [v0.1.1 安装记录](evidence/releases/v0.1.1/INSTALLATION_VERIFICATION.zh-CN.md)。
 
 ## 安装范围
 

docs/MAINTENANCE.zh-CN.md

@@ -14,10 +14,10 @@
 
 新增、更新、改名、弃用或移除时：
 
-1. 先执行 [Skill 准入](SKILL_ADMISSION.md) 的 reuse-before-invention 与 ownership gate。
+1. 先执行 [Skill 准入](SKILL_ADMISSION.zh-CN.md) 的 reuse-before-invention 与 ownership gate。
 2. 保留包边界、invocation direction、attribution 和必要资源。
 3. 收集 structural、fresh-install、behavioral、invocation、script 和 review evidence，并准确标注 evidence class。
-4. 执行 [review policy](REVIEW_POLICY.md) 的 review trigger。
+4. 执行[审查策略](REVIEW_POLICY.zh-CN.md)的 review trigger。
 5. 同步 README、目录、安装说明、治理链接、affected recipes、discovery tests、attribution 和 changelog。
 6. 未通过真实 release gate 前，不把版本、tag 或安装命令写成已验证。
 

docs/evidence/releases/v0.1.1/RELEASE_RECEIPT.zh-CN.md

@@ -19,15 +19,15 @@ Actions 和 fresh-install 证据已核验；独立 `review-loop agent-skill` acc
 
 ## 证据
 
-- 结构与 discovery：[DISCOVERY_VERIFICATION.md](DISCOVERY_VERIFICATION.md)
-- 包与集合测试：[TEST_SUMMARY.md](TEST_SUMMARY.md)
-- Fresh install 与 host discovery：[INSTALLATION_VERIFICATION.md](INSTALLATION_VERIFICATION.md)
-- 限制和 evidence labels：[LIMITATIONS.md](LIMITATIONS.md)
-- [审查策略](../../../REVIEW_POLICY.md)
-- [准入契约](../../../SKILL_ADMISSION.md)
-
-| Fresh whole-repository install | `VERIFIED` | 见 [INSTALLATION_VERIFICATION.md](INSTALLATION_VERIFICATION.md)。 |
-| Fresh per-Skill install | `VERIFIED` | 见 [INSTALLATION_VERIFICATION.md](INSTALLATION_VERIFICATION.md)。 |
+- 结构与 discovery：[DISCOVERY_VERIFICATION.zh-CN.md](DISCOVERY_VERIFICATION.zh-CN.md)
+- 包与集合测试：[TEST_SUMMARY.zh-CN.md](TEST_SUMMARY.zh-CN.md)
+- Fresh install 与 host discovery：[INSTALLATION_VERIFICATION.zh-CN.md](INSTALLATION_VERIFICATION.zh-CN.md)
+- 限制和 evidence labels：[LIMITATIONS.zh-CN.md](LIMITATIONS.zh-CN.md)
+- [审查策略](../../../REVIEW_POLICY.zh-CN.md)
+- [准入契约](../../../SKILL_ADMISSION.zh-CN.md)
+
+| Fresh whole-repository install | `VERIFIED` | 见 [INSTALLATION_VERIFICATION.zh-CN.md](INSTALLATION_VERIFICATION.zh-CN.md)。 |
+| Fresh per-Skill install | `VERIFIED` | 见 [INSTALLATION_VERIFICATION.zh-CN.md](INSTALLATION_VERIFICATION.zh-CN.md)。 |
 | Independent `review-loop agent-skill` acceptance | `BLOCKED` | 等待 fresh independent evaluator 记录；同一上下文不是独立证据。 |
 | GitHub Actions | `VERIFIED` | 合并后的 release commit 上的 workflow 已通过。 |
 

docs/evidence/releases/v0.1.1/TEST_SUMMARY.zh-CN.md

@@ -17,7 +17,7 @@
 | Quick Start | `powershell -File tests/quick-start-smoke-tests.ps1` | `PASS`；8 条断言。 |
 | Python 语法 | 只读 `ast.parse` 检查 | `PASS`；12 个 Python 文件解析通过且不写入字节码。 |
 | Python `compileall` | `python -m compileall -q skills tests` | 本地 `NOT PASSED`；沙箱拒绝写入现有受保护的 `__pycache__`，但合并 release commit `c50f1ef` 上的 GitHub Actions 同命令已通过。 |
-| Tagged fresh install | `npx skills add LightDevCoder/skills#v0.1.1 ...` 加 `npx skills list` | `PASS`；CLI `1.5.20`，整仓 destination 列出 5 个包，单 Skill destination 列出 `review-loop`，且两个 source checkout 都不存在。见[安装证据](INSTALLATION_VERIFICATION.md)。 |
+| Tagged fresh install | `npx skills add LightDevCoder/skills#v0.1.1 ...` 加 `npx skills list` | `PASS`；CLI `1.5.20`，整仓 destination 列出 5 个包，单 Skill destination 列出 `review-loop`，且两个 source checkout 都不存在。见[安装证据](INSTALLATION_VERIFICATION.zh-CN.md)。 |
 | Release-commit CI | GitHub Actions `collection-quality` | `PASS`；run `30189210521` 对应 `c50f1ef403a5f0bfe02e75d1aeff2c237556db63`。 |
 
 包测试只证明覆盖到的 contract 场景；collection discovery 只证明 cross-reference 和 metadata 一致。tagged installation 证据另有记录；这些证据都不能证明 independent acceptance 或模型介导的 runtime 行为。

docs/skills/manuscript-ops.md

@@ -38,8 +38,10 @@ records, review evidence, and production QA records.
 ## Handoffs, success, and `BLOCKED`
 
 On a Project route, choose one discovery handoff: `grill-me` for one-session
-decisions or `wayfinder` for multi-session uncertainty, then stop. Continue
-only after the user explicitly resumes `manuscript-ops`. Before initialization,
+decisions or `wayfinder` for multi-session uncertainty, then stop. `grill-me`
+is the user-facing entry point; its underlying model-invoked capability is
+`grilling`, so it is one clarification capability rather than two user steps.
+Continue only after the user explicitly resumes `manuscript-ops`. Before initialization,
 the user must explicitly choose `project-init`; before review, an approved
 `review-loop` Charter is required. `BLOCKED` is correct for a missing root,
 missing dependency, unapproved brief, absent capability, missing evidence, or

docs/skills/project-init.md

@@ -25,8 +25,9 @@ defaults to the current directory. Before writing, it reads root
 `AGENTS.md`/`CLAUDE.md`, README, manifests, project documents, and current
 status. It needs the project type, visible goal, outputs, collaboration mode,
 constraints, and required review level, either from the brief or from its
-short questions. `grilling` may help answer missing questions; it does not
-authorize additional writes.
+short questions. Use the external user-facing `grill-me` entry point when a
+question needs a clarification interview; its underlying model-invoked
+capability is `grilling`. This does not authorize additional writes.
 
 ```text
 $project-init