qxbyte · qxbyte · Jul 6, 2026 · Jul 6, 2026
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
@@ -10,7 +10,7 @@
     {
       "name": "specode",
       "source": "./plugins/specode",
-      "version": "6.1.4",
+      "version": "6.1.5",
       "description": "Lightweight spec-driven workflow orchestration shell — walks a host agent through requirements → design → tasks → execute → acceptance, delegating each phase to superpowers skills with a first-class specode-native fallback, and landing 4 fixed docs per spec (requirements / design / tasks / implementation-log). Bundles a dedicated intake skill for the requirements phase, a zero-import task-swarm handoff for parallel execution, and optional locate-oriented experience retrieval. Version history is in the CHANGELOG.",
       "category": "development",
       "homepage": "https://github.com/qxbyte/pluginhub",
@@ -25,7 +25,7 @@
     {
       "name": "task-swarm",
       "source": "./plugins/task-swarm",
-      "version": "0.10.3",
+      "version": "0.10.4",
       "description": "Standalone multi-agent orchestration driven by a pipeline.yml — semantic task groups with cross-group concurrency, forked coders, and per-group reviewer + validator loops (state.json is the single source of truth). specode delegates its execution phase here; also runnable directly via /task-swarm:swarm. Version history is in the CHANGELOG.",
       "category": "development",
       "homepage": "https://github.com/qxbyte/pluginhub",
@@ -41,7 +41,7 @@
     {
       "name": "obsidian-wiki",
       "source": "./plugins/obsidian-wiki",
-      "version": "2.0.3",
+      "version": "2.0.4",
       "description": "Maintain an Obsidian LLM-Wiki via three skills — a deterministic structure layer (wiki-struct: Home tree / per-dir READMEs / partition pages), content curation (wiki-curate: ingest / curate / lint), and a unified orchestrator (wiki-orchestrate). Generic code + per-vault config in the home-dir registry ~/.config/obsidian-wiki/ (fallback: <vault>/.wiki/config.json), zero hardcoded structure. Version history is in the CHANGELOG.",
       "category": "productivity",
       "homepage": "https://github.com/qxbyte/pluginhub",
@@ -58,7 +58,7 @@
     {
       "name": "ragkit",
       "source": "./plugins/ragkit",
-      "version": "0.1.1",
+      "version": "0.1.2",
       "description": "Standalone knowledge-base RAG plugin (vector/lexical/metadata multi-channel recall).",
       "category": "database",
       "homepage": "https://github.com/qxbyte/pluginhub",

diff --git a/plugins/obsidian-wiki/.claude-plugin/plugin.json b/plugins/obsidian-wiki/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "obsidian-wiki",
-  "version": "2.0.3",
+  "version": "2.0.4",
   "description": "Maintain an Obsidian LLM-Wiki via three skills — a deterministic structure layer (wiki-struct: Home tree / per-dir READMEs / partition pages), content curation (wiki-curate: ingest / curate / lint), and a unified orchestrator (wiki-orchestrate). Generic code + per-vault config in the home-dir registry ~/.config/obsidian-wiki/ (fallback: <vault>/.wiki/config.json), zero hardcoded structure. Version history is in the CHANGELOG.",
   "author": {
     "name": "xueqiang",

diff --git a/plugins/obsidian-wiki/CHANGELOG.md b/plugins/obsidian-wiki/CHANGELOG.md
@@ -4,6 +4,10 @@ obsidian-wiki 是维护 Obsidian LLM-Wiki 的多 agent 插件（从独立 skills
 
 ## Unreleased
 
+## 2.0.4 (2026-07-06) — skill description 统一模板
+
+- 三个 skill（wiki-struct / wiki-curate / wiki-orchestrate）的 frontmatter `description`（渐进式加载唯一常驻的元数据）统一为「用途开头 → 做什么 → 触发语 → 边界/交叉引用」轻量模板：补上彼此的交叉引用边界（结构层→wiki-struct、内容策展→wiki-curate），删去版本史噪音（如 wiki-curate 的「v2.0.0 剥离」）。零行为变化。
+
 ## 2.0.3 (2026-07-05) — Discover 分类标签
 
 - marketplace.json 加 `"category": "productivity"`，Discover 面板显示 `[productivity]` 标签。零行为变化。

diff --git a/plugins/obsidian-wiki/skills/wiki-curate/SKILL.md b/plugins/obsidian-wiki/skills/wiki-curate/SKILL.md
@@ -1,6 +1,6 @@
 ---
 name: wiki-curate
-description: 方法论伞 + ingest/curate/lint（内容向）——基于 Karpathy LLM Wiki 三层架构，对 vault 执行 Scan → Ingest → Curate → Lint 四阶段内容策展。结构层（Home 树/分区页/地图）见 wiki-struct；SpecIn 知识沉淀已迁移到 specode 的 /specode:distill（v2.0.0 剥离）。绝不修改只读目录原文（07-Ideas/、10-Work/、SpecIn/），不写 wiki-struct 受管块，不写遗留的 10-Work/知识库/ 产物。触发语：/wiki-curate、scan、ingest、curate、lint、「整理我的笔记库」、「整理 vault」、「整理 LLM Wiki」。
+description: 用于对 vault 做内容策展——基于 Karpathy LLM Wiki 三层架构执行 Scan → Ingest → Curate → Lint 四阶段。触发语：/wiki-curate、scan、ingest、curate、lint、「整理我的笔记库」、「整理 vault」、「整理 LLM Wiki」。绝不改只读目录原文（07-Ideas/、10-Work/、SpecIn/），不写 wiki-struct 受管块；结构层（Home 树/分区页/地图）→ wiki-struct。
 ---
 
 

diff --git a/plugins/obsidian-wiki/skills/wiki-orchestrate/SKILL.md b/plugins/obsidian-wiki/skills/wiki-orchestrate/SKILL.md
@@ -1,13 +1,12 @@
 ---
 name: wiki-orchestrate
 description: >
-  整理 Obsidian LLM Wiki 的统一编排入口——定位 vault → 两方只读体检 →
-  汇总行动计划 → 用户确认 → 按「结构→策展」编排调用 wiki-struct、
-  wiki-curate 完成整理。下一步该跑哪个 skill / 命令 / 参数由
-  模型依据体检结果判断（建议性，非规则）。触发语：/wiki-orchestrate、
-  /wiki-orchestrate scan、/wiki-orchestrate set-vault、「整理笔记库」、
-  「跑一次编排」、「统一整理 vault」。入口自身不直接改内容笔记，所有写操作
-  委托给子 skill 并继承其红线。
+  用于整理 Obsidian LLM Wiki 的统一编排入口——定位 vault → 两方只读体检 →
+  汇总行动计划 → 用户确认 → 按「结构→策展」编排 wiki-struct / wiki-curate
+  完成整理（下一步跑哪个 skill / 命令 / 参数由模型依据体检结果判断，建议性非规则）。
+  触发语：/wiki-orchestrate（scan / set-vault）、「整理笔记库」、
+  「跑一次编排」、「统一整理 vault」。入口自身不改内容笔记，写操作全委托子
+  skill 并继承其红线。
 ---
 
 # wiki-orchestrate

diff --git a/plugins/obsidian-wiki/skills/wiki-struct/SKILL.md b/plugins/obsidian-wiki/skills/wiki-struct/SKILL.md
@@ -1,6 +1,6 @@
 ---
 name: wiki-struct
-description: 维护 Obsidian LLM Wiki 的"结构层"——确定性重写 Home 总览树、各一级目录 README、00-Index 分区页的"受管块"，并产出结构体检报告。触发语：/wiki-struct、/wiki-struct check、/wiki-struct apply、/wiki-struct init、"刷新 Home 树"、"更新 README 目录"、"整理结构层"。绝不修改内容笔记原文，只重写 marker 包裹的受管块。
+description: 用于维护 Obsidian LLM Wiki 的「结构层」——确定性重写 Home 总览树、各一级目录 README、00-Index 分区页的「受管块」，并产出结构体检报告。触发语：/wiki-struct（check / apply / init）、「刷新 Home 树」、「更新 README 目录」、「整理结构层」。只重写 marker 包裹的受管块，绝不改内容笔记原文；内容策展 → wiki-curate。
 ---
 
 # Wiki Struct

diff --git a/plugins/ragkit/.claude-plugin/plugin.json b/plugins/ragkit/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "ragkit",
-  "version": "0.1.1",
+  "version": "0.1.2",
   "description": "Standalone knowledge-base RAG: chunk + embed + multi-channel recall (vector/lexical/metadata, RRF-fused) over distill-style knowledge-base/ directories.",
   "author": {
     "name": "xueqiang",

diff --git a/plugins/ragkit/CHANGELOG.md b/plugins/ragkit/CHANGELOG.md
@@ -2,6 +2,10 @@
 
 ## Unreleased
 
+## 0.1.2 (2026-07-06) — skill description 统一模板
+
+- 四个 skill（embed / query / status / eval）的 frontmatter `description`（渐进式加载唯一常驻的元数据）统一为「Use when 场景 → 做什么 → Trigger 触发词/命令 → 边界」轻量模板：补上 `/ragkit:*` 触发命令与彼此的交叉引用边界（query↔embed↔status）。零行为变化。
+
 ## 0.1.1 (2026-07-05) — Discover 分类标签
 
 - marketplace.json 加 `"category": "database"`，Discover 面板显示 `[database]` 标签。零行为变化。

diff --git a/plugins/ragkit/skills/embed/SKILL.md b/plugins/ragkit/skills/embed/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: embed
 user-invocable: false
-description: Use when building or refreshing the RagKit vector index for a project knowledge-base (after distill produced/updated knowledge points)
+description: Use when building or refreshing the RagKit index for a project `knowledge-base/` (e.g. after distill added/updated knowledge points). Trigger: /ragkit:embed. Querying instead → ragkit:query; index health → ragkit:status.
 ---
 
 # RagKit Embed

diff --git a/plugins/ragkit/skills/eval/SKILL.md b/plugins/ragkit/skills/eval/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: eval
 user-invocable: false
-description: Use when measuring or tuning RagKit retrieval accuracy — runs the golden evalset and reports recall@k / MRR per bucket
+description: Use when measuring or tuning RagKit retrieval accuracy — runs the golden evalset, reports recall@k / MRR per bucket. Trigger: /ragkit:eval. Run before/after any retrieval-param change to keep a baseline.
 ---
 
 # RagKit Eval

diff --git a/plugins/ragkit/skills/query/SKILL.md b/plugins/ragkit/skills/query/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: query
 user-invocable: false
-description: Use when the user (or a workflow like specode) needs to retrieve from a project knowledge-base — multi-channel RAG recall returning pointer cards, supports multi-round multi-angle querying
+description: Use when retrieving from a project `knowledge-base/` — multi-channel RAG recall (vector + lexical + metadata, RRF-fused) returning locate-pointer cards; multi-round/multi-angle querying allowed. Trigger: /ragkit:query, or called by name by workflows like specode. Building/refreshing the index instead → ragkit:embed.
 ---
 
 # RagKit Query

diff --git a/plugins/ragkit/skills/status/SKILL.md b/plugins/ragkit/skills/status/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: status
 user-invocable: false
-description: Use when checking RagKit index health — doc/chunk counts, backend resolution, staleness and drift between knowledge-base and index
+description: Use when checking RagKit index health — doc/chunk counts, backend resolution, staleness and drift between `knowledge-base/` and index. Trigger: /ragkit:status. Rebuilding after drift → ragkit:embed.
 ---
 
 # RagKit Status

diff --git a/plugins/specode/.claude-plugin/plugin.json b/plugins/specode/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "specode",
-  "version": "6.1.4",
+  "version": "6.1.5",
   "description": "Lightweight spec-driven workflow orchestration shell — walks a host agent through requirements → design → tasks → execute → acceptance, delegating each phase to superpowers skills with a first-class specode-native fallback, and landing 4 fixed docs per spec (requirements / design / tasks / implementation-log). Bundles a dedicated intake skill for the requirements phase, a zero-import task-swarm handoff for parallel execution, and optional locate-oriented experience retrieval. Version history is in the CHANGELOG.",
   "author": {
     "name": "xueqiang",

diff --git a/plugins/specode/CHANGELOG.md b/plugins/specode/CHANGELOG.md
@@ -4,6 +4,10 @@ specode 是 spec-driven 轻量工作流插件：requirements → design → task
 
 ## Unreleased
 
+## 6.1.5 (2026-07-06) — skill description 统一模板
+
+- **skill `description` 统一为渐进式加载模板（零行为变化）**：三个 skill（specode / intake / distill）的 frontmatter `description` 是 agent 渐进式加载时**唯一常驻**的元数据，模型据此决定何时唤起该 skill。统一改写为「用途场景开头（Use when …）→ 做什么/产物 → Trigger 触发词/命令 → 边界/谁触发」的轻量模板，并删去描述里的版本史噪音（如 intake 的「replaces the old brainstorming…」）。正文与行为逻辑一字未动。
+
 ## 6.1.4 (2026-07-05)
 
 - **marketplace 加 `category` 字段**：在 `.claude-plugin/marketplace.json` 的 specode 条目加 `"category": "development"`，让 `/plugin` Discover 面板在插件名后显示 `[development]` 标签（对齐官方市场；该字段只在 marketplace.json、UI 只读它）。零行为变化。

diff --git a/plugins/specode/skills/distill/SKILL.md b/plugins/specode/skills/distill/SKILL.md
@@ -2,13 +2,11 @@
 name: distill
 user-invocable: false
 description: >
-  Manually distill a single specode-managed spec into atomic, locate-oriented
-  markdown knowledge points. Default output: md-only, written to the project's
-  own `<project_root>/knowledge-base/` (cases/ + navigation/ + MEMORY.md);
-  optionally copied to an Obsidian vault. Trigger ONLY via
-  `/specode:distill <slug>` — never auto-run by the main specode flow.
-  Md-only, no auto-injection. Pure "atomic locate-pointer sedimenter for the
-  current project".
+  Use when distilling a finished specode spec into atomic, locate-oriented
+  markdown knowledge points — writes cases/ + navigation/ + MEMORY.md into the
+  project's own `<project_root>/knowledge-base/` (md-only, no auto-injection),
+  optionally copied to an Obsidian vault. Trigger: ONLY `/specode:distill <slug>`
+  — never auto-run by the main specode flow.
 ---
 
 # distill — project knowledge-base sedimenter (atomic dual-track)

diff --git a/plugins/specode/skills/intake/SKILL.md b/plugins/specode/skills/intake/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: intake
 user-invocable: false
-description: specode's requirements-intake engine — project analysis + code location + clarification + writing requirements.md. Invoked by name (specode:intake) via the Skill tool by the specode orchestration SKILL at the start of a spec (requirements phase); it replaces the old "brainstorming moonlighting as a requirements writer / thin native questionnaire". Produces <specsRoot>/<slug>/requirements.md (with the project_root frontmatter contract) plus the location pointers handed to the design phase. Not a user-triggered command.
+description: Use when running specode's requirements phase — project analysis + code location + clarification, then writing requirements.md. Produces <specsRoot>/<slug>/requirements.md (with the project_root frontmatter contract) plus the location pointers handed to the design phase. Invoked by name (specode:intake) via the Skill tool by the specode orchestration SKILL; not a user-triggered command.
 ---
 
 # intake — specode's requirements-intake engine

diff --git a/plugins/specode/skills/specode/SKILL.md b/plugins/specode/skills/specode/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: specode
 user-invocable: false
-description: Lightweight spec-driven workflow orchestration shell. Across the requirements → design → tasks → 「执行方式」 → execute → acceptance phases it autonomously calls mature superpowers skills to do the heavy lifting (clarification, design, planning, TDD execution, acceptance), falling back to specode-native when superpowers is absent, and files the four fixed artifacts (requirements.md / design.md / tasks.md / implementation-log.md) into the user's document directory. Activates only when the user invokes `/specode:spec <request>`, `/specode:continue <slug>`, `/specode:list`, or explicitly asks to enter spec mode; otherwise behave as a normal conversation.
+description: Use when running a spec-driven workflow — orchestration shell that walks requirements → design → tasks → 「执行方式」 → execute → acceptance, autonomously calling superpowers skills for the heavy lifting (clarification / design / planning / TDD execution / acceptance) with a first-class specode-native fallback, and filing 4 fixed artifacts (requirements.md / design.md / tasks.md / implementation-log.md) under the user's document directory. Trigger: /specode:spec <request>, /specode:continue <slug>, /specode:list, or "进入 spec 模式". Outside these, behave as a normal conversation.
 ---
 
 # specode — orchestration shell

diff --git a/plugins/task-swarm/.claude-plugin/plugin.json b/plugins/task-swarm/.claude-plugin/plugin.json
@@ -1,6 +1,6 @@
 {
   "name": "task-swarm",
-  "version": "0.10.3",
+  "version": "0.10.4",
   "description": "Standalone multi-agent orchestration driven by a pipeline.yml — semantic task groups with cross-group concurrency, forked coders, and per-group reviewer + validator loops (state.json is the single source of truth). specode delegates its execution phase here; also runnable directly via /task-swarm:swarm. Version history is in the CHANGELOG.",
   "author": {
     "name": "xueqiang",

diff --git a/plugins/task-swarm/CHANGELOG.md b/plugins/task-swarm/CHANGELOG.md
@@ -4,6 +4,10 @@ task-swarm 是从 specode 拆出的独立多 agent 编排 plugin。本文件记
 
 ## Unreleased
 
+## 0.10.4 (2026-07-06) — skill description 统一模板
+
+- skill `description`（渐进式加载唯一常驻的元数据）统一为「Use when 场景 → 做什么 → Trigger 触发词/命令 → 边界/谁触发」轻量模板，`Trigger words —` 标签规范为 `Trigger:`，并补上「specode 执行阶段委托至此」的触发场景。零行为变化。
+
 ## 0.10.3 (2026-07-05) — Discover 分类标签
 
 - marketplace.json 加 `"category": "development"`，Discover 面板显示 `[development]` 标签。零行为变化。

diff --git a/plugins/task-swarm/skills/task-swarm/SKILL.md b/plugins/task-swarm/skills/task-swarm/SKILL.md
@@ -1,7 +1,7 @@
 ---
 name: task-swarm
 user-invocable: false
-description: Use when driving the task-swarm multi-agent orchestrator standalone — reading a requirement/design doc, generating a pipeline.yml, then running the fork → review → validate loop to complete a multi-task implementation. Trigger words — task-swarm, concurrent task execution, pipeline.yml orchestration, multi-coder fork.
+description: Use when driving the task-swarm multi-agent orchestrator standalone — read a requirement/design doc, generate a pipeline.yml, then run the fork → review → validate loop to finish a multi-task implementation. Trigger: /task-swarm:swarm, "task-swarm", concurrent task execution, pipeline.yml orchestration, multi-coder fork; also invoked when specode's execution phase delegates here.
 ---
 
 # task-swarm standalone orchestration SKILL