chore: prepare v0.7.2 release

Author: SummerSec

.github/workflows/release.yml

@@ -23,19 +23,29 @@ jobs:
         run: |
           VERSION="${{ steps.version.outputs.VERSION }}"
           # Extract section between current version and next (or end)
-          CHANGELOG=$(awk -v ver="$VERSION" '
+          CHANGELOG="$(awk -v ver="$VERSION" '
             /^## \[/ { in_section = 0 }
             index($0, "## [" ver "]") == 1 { in_section = 1; next }
             in_section && /^## \[/ { exit }
             in_section { print }
-          ' CHANGELOG.md | tail -n +2)
-          echo "BODY=${CHANGELOG//$'\n'/\\n}" >> $GITHUB_OUTPUT
+          ' CHANGELOG.md | sed '/./,$!d')"
+
+          if [ -z "$CHANGELOG" ]; then
+            echo "No CHANGELOG section found for version $VERSION" >&2
+            exit 1
+          fi
+
+          {
+            echo "BODY<<EOF"
+            echo "$CHANGELOG"
+            echo "EOF"
+          } >> "$GITHUB_OUTPUT"
 
       - name: Create GitHub Release
         uses: softprops/action-gh-release@v2
         with:
           tag_name: ${{ github.ref }}
-          name: ${{ steps.version.outputs.VERSION }}
+          name: v${{ steps.version.outputs.VERSION }}
           body: ${{ steps.changelog.outputs.BODY }}
           draft: false
           prerelease: false

CHANGELOG.md

@@ -12,6 +12,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Claude Code plugin**: Store session state and active persona under `${CLAUDE_PLUGIN_DATA}` so marketplace installs keep persistent data across plugin cache updates.
 - **Persona switching**: Avoid modifying plugin cache source files when `/inner-os persona use` runs inside Claude Code; repository and global-install workflows still update platform adapter files.
 
+### Added
+
+- **Claude Code hooks**: Add `PostCompact`, `SubagentStart`, and `SubagentStop` coverage to preserve Inner OS continuity after compaction and during subagent lifecycles.
+- **Cursor plugin packaging**: Add `.cursor-plugin/` metadata so the existing `cursor/` adapter can be discovered as a Cursor plugin component directory.
+- **OpenClaw extension packaging**: Add OpenClaw plugin metadata and extension entrypoint for JSONL capture of `▎InnerOS：` outputs.
+- **Documentation**: Document Claude Code and Cursor plugin packaging standards in `CLAUDE.md`, including persistent data rules and component path constraints.
+
 ## [0.7.1] - 2026-04-16
 
 ### Fixed

CLAUDE.md

@@ -34,6 +34,8 @@ PostToolUse (success) → normalizes event → appends to state → injects rece
 PostToolUseFailure    → normalizes failure → increments failureCount → injects error context
      ↓
 PreCompact → saves compactedAt timestamp
+PostCompact → injects compact-continuity context with recent events
+SubagentStart/SubagentStop → records subagent lifecycle events
      ↓
 Stop → deletes session state file
 ```
@@ -67,7 +69,7 @@ Platforms degrade gracefully in hook richness:
 
 | Platform | Protocol mechanism | Hook scripts | Reuses hooks/lib/ |
 |----------|-------------------|--------------|-------------------|
-| Claude Code | Dynamic (reads SKILL.md) | 6 hooks | Yes (canonical) |
+| Claude Code | Dynamic (reads SKILL.md) | 9 hooks | Yes (canonical) |
 | Codex CLI | SessionStart + PostToolUse + Stop | 3 hooks | Yes |
 | Cursor | sessionStart + postToolUse + stop | 3 hooks | Yes |
 | OpenCode | Plugin + static instructions | Plugin | No |
@@ -113,7 +115,7 @@ This repository also ships Cursor plugin metadata because AI Inner OS is multi-p
 - Every hook wraps its body in `try/catch` and fails silently — hook errors never interrupt the session
 - Claude Code plugin session state files live in `${CLAUDE_PLUGIN_DATA}/state/`; repo/global-install fallback state files live in `state/` (gitignored), keyed by sanitized session ID
 - `failureCount` increments on consecutive failures, resets to 0 on any success
-- Claude Code: PreToolUse uses `hookSpecificOutput.additionalContext`; PostToolUse/PostToolUseFailure output plain text to stdout
+- Claude Code: PreToolUse uses `hookSpecificOutput.additionalContext`; PostToolUse/PostToolUseFailure/PostCompact/SubagentStart/SubagentStop output plain text to stdout
 - Cursor: sessionStart/postToolUse use `{ additional_context: string }` top-level format; preToolUse removed (can't inject context)
 - Codex: SessionStart outputs plain text to stdout; PostToolUse uses `hookSpecificOutput.additionalContext` JSON; PreToolUse removed (additionalContext not supported)
 - Bash commands are truncated at 80 chars in target extraction

README.md

@@ -203,6 +203,8 @@ Core principles:
 | Protocol Injection | Hook reads SKILL.md dynamically | SessionStart Hook | sessionStart Hook | Plugin + instructions | Skill or `.hermes.md` | Skill (AgentSkills format) |
 | Post-tool hook | `PostToolUse` | `PostToolUse` | `postToolUse` | Plugin event | — | — |
 | Failure tracking | `PostToolUseFailure` | — | — | — | — | — |
+| Compaction continuity | `PreCompact` + `PostCompact` | — | — | — | — | — |
+| Subagent lifecycle | `SubagentStart` + `SubagentStop` | — | — | — | — | — |
 | Persona switching | `/inner-os persona` command | Dynamic (Hook reads) | Dynamic (Hook reads) | Plugin tool | Script injection | Script injection |
 | Installation | Plugin marketplace one-click | `install.js` global | `install.js` global | `install.js` global | `install.js` global | `install.js` global |
 | Shared logic | `hooks/lib/` (canonical) | Reuses `hooks/lib/` | Reuses `hooks/lib/` | Standalone Plugin | Static injection only | Static injection only |
@@ -219,6 +221,10 @@ PreToolUse → Tool execution → PostToolUse (success)
                  ↓
 PreCompact → Save state
                  ↓
+PostCompact → Restore compact-continuity context
+                 ↓
+SubagentStart/SubagentStop → Track subagent lifecycle
+                 ↓
 Stop → Clean up state
 ```
 
@@ -229,6 +235,9 @@ Stop → Clean up state
 | `PostToolUse` | After successful execution | Track events, inject recent activity context |
 | `PostToolUseFailure` | After failed execution | Track failures, inject error context and consecutive failure count |
 | `PreCompact` | Before context compaction | Save state, maintain protocol continuity |
+| `PostCompact` | After context compaction | Inject compact-continuity context |
+| `SubagentStart` | Subagent starts | Track subagent lifecycle start |
+| `SubagentStop` | Subagent stops | Track subagent lifecycle completion |
 | `Stop` | Session end | Clean up state files |
 
 ---

README_CN.md

@@ -203,6 +203,8 @@ Inner OS 的行为协议定义在 [`protocol/SKILL.md`](protocol/SKILL.md)，是
 | 协议注入 | Hook 动态读取 SKILL.md | SessionStart Hook | sessionStart Hook | Plugin + instructions | Skill 或 `.hermes.md` | Skill（AgentSkills 格式） |
 | 工具执行后 hook | `PostToolUse` | `PostToolUse` | `postToolUse` | Plugin event | — | — |
 | 失败追踪 | `PostToolUseFailure` | — | — | — | — | — |
+| 压缩连续性 | `PreCompact` + `PostCompact` | — | — | — | — | — |
+| 子代理生命周期 | `SubagentStart` + `SubagentStop` | — | — | — | — | — |
 | 人设切换 | `/inner-os persona` 命令 | 动态（Hook 读取） | 动态（Hook 读取） | Plugin tool | 脚本注入 | 脚本注入 |
 | 安装方式 | 插件市场一键安装 | `install.js` 全局安装 | `install.js` 全局安装 | `install.js` 全局安装 | `install.js` 全局安装 | `install.js` 全局安装 |
 | 共享逻辑 | `hooks/lib/`（原始实现） | 复用 `hooks/lib/` | 复用 `hooks/lib/` | 独立 Plugin | 纯静态注入 | 纯静态注入 |
@@ -219,6 +221,10 @@ PreToolUse → 工具执行 → PostToolUse (成功)
                  ↓
 PreCompact → 保存状态
                  ↓
+PostCompact → 恢复压缩后的连续上下文
+                 ↓
+SubagentStart/SubagentStop → 追踪子代理生命周期
+                 ↓
 Stop → 清理状态
 ```
 
@@ -229,6 +235,9 @@ Stop → 清理状态
 | `PostToolUse` | 工具执行成功后 | 追踪事件，注入最近活动上下文 |
 | `PostToolUseFailure` | 工具执行失败后 | 追踪失败，注入错误上下文和连续失败计数 |
 | `PreCompact` | 上下文压缩前 | 保存状态，维持协议连续性 |
+| `PostCompact` | 上下文压缩后 | 注入压缩后连续性上下文 |
+| `SubagentStart` | 子代理启动 | 记录子代理开始事件 |
+| `SubagentStop` | 子代理结束 | 记录子代理结束事件 |
 | `Stop` | 会话结束 | 清理状态文件 |
 
 ---

codex/README.md

@@ -67,5 +67,5 @@ Codex CLI 支持读取 Inner OS 人设配置。手动编辑 `personas/_active.js
 |---|---|---|
 | 协议注入 | SessionStart hook 自动读取 SKILL.md | AGENTS.md 静态加载 |
 | 失败追踪 | PostToolUseFailure 独立 hook | 无（Codex 暂不支持） |
-| Hook 数量 | 6 个 | 4 个 |
+| Hook 数量 | 9 个 | 3 个 |
 | 安装方式 | 插件市场一键安装 | 手动复制配置文件 |

cursor/README.md

@@ -74,6 +74,6 @@ node ~/.inner-os/scripts/switch-persona.js default
 | | Claude Code | Codex CLI | Cursor |
 |---|---|---|---|
 | 协议注入 | SessionStart hook 读取 SKILL.md | SessionStart hook / AGENTS.md | `.mdc` 规则 + `sessionStart` hook |
-| Hook 数量 | 6 个 | 3 个 | 3 个 |
+| Hook 数量 | 9 个 | 3 个 | 3 个 |
 | 失败追踪 | PostToolUseFailure | 不支持 | 不支持 |
 | 安装方式 | 插件市场一键安装 | `install.js` 全局安装 | Cursor 插件清单 / `install.js` 全局安装 |

docs/install-claude-code.md

@@ -1,6 +1,6 @@
 # Claude Code 安装指南
 
-在 [Claude Code](https://claude.ai/code) 中安装 AI Inner OS。这是功能最完整的平台，拥有 6 个生命周期 hook。
+在 [Claude Code](https://claude.ai/code) 中安装 AI Inner OS。这是功能最完整的平台，拥有 9 个生命周期 hook。
 
 ## 前置条件
 
@@ -80,6 +80,10 @@ PostToolUseFailure (失败) → 追踪失败，注入错误上下文
        ↓
 PreCompact → 保存状态
        ↓
+PostCompact → 恢复压缩后的连续上下文
+       ↓
+SubagentStart/SubagentStop → 追踪子代理生命周期
+       ↓
 Stop → 清理状态
 ```
 
@@ -90,6 +94,9 @@ Stop → 清理状态
 | `PostToolUse` | 工具执行成功后 | 追踪事件，注入最近活动上下文 |
 | `PostToolUseFailure` | 工具执行失败后 | 追踪失败，注入错误上下文和连续失败计数 |
 | `PreCompact` | 上下文压缩前 | 保存状态，维持协议连续性 |
+| `PostCompact` | 上下文压缩后 | 注入压缩后连续性上下文 |
+| `SubagentStart` | 子代理启动 | 记录子代理开始事件 |
+| `SubagentStop` | 子代理结束 | 记录子代理结束事件 |
 | `Stop` | 会话结束 | 清理状态文件 |
 
 ## 文件说明
@@ -102,6 +109,9 @@ Stop → 清理状态
 | `hooks/post-tool-use.js` | 归一化事件，更新状态 |
 | `hooks/post-tool-use-failure.js` | 追踪失败事件 |
 | `hooks/pre-compact.js` | 压缩前保存状态 |
+| `hooks/post-compact.js` | 压缩后注入连续性上下文 |
+| `hooks/subagent-start.js` | 记录子代理启动事件 |
+| `hooks/subagent-stop.js` | 记录子代理结束事件 |
 | `hooks/stop.js` | 清理会话状态文件 |
 | `hooks/lib/` | 共享逻辑（状态管理、事件归一化、prompt 拼装等） |
 | `protocol/SKILL.md` | Inner OS 行为协议（唯一数据源） |

docs/install-codex.md

@@ -81,7 +81,7 @@ codex_hooks = true
 |---|---|---|
 | 协议注入 | Hook 动态读取 `SKILL.md` | `SessionStart` Hook + `AGENTS.md` |
 | 失败追踪 | `PostToolUseFailure` 独立 hook | 不支持 |
-| Hook 数量 | 6 个 | 3 个 |
+| Hook 数量 | 9 个 | 3 个 |
 | 工具覆盖 | 所有工具 | 仅 Bash |
 | 安装方式 | 插件市场一键安装 | 全局安装脚本或手动 |
 | 路径解析 | `${CLAUDE_PLUGIN_ROOT}` 自动解析 | 安装脚本生成绝对路径 |

docs/installation.md

@@ -13,7 +13,7 @@ AI Inner OS 支持 6 个平台。选择你使用的平台查看详细安装文
 
 ## 快速选择
 
-- **想要最完整的体验？** → [Claude Code](install-claude-code.md)（6 个生命周期 hook，插件市场一键安装）
+- **想要最完整的体验？** → [Claude Code](install-claude-code.md)（9 个生命周期 hook，插件市场一键安装）
 - **使用 OpenAI 生态？** → [Codex CLI](install-codex.md)（4 个 hook，手动配置）
 - **用 IDE 写代码？** → [Cursor](install-cursor.md)（规则文件自动生效）
 - **轻量级 CLI？** → [OpenCode CLI](install-opencode.md)（纯静态注入）
@@ -24,7 +24,7 @@ AI Inner OS 支持 6 个平台。选择你使用的平台查看详细安装文
 | | Claude Code | Codex CLI | Cursor | OpenCode | Hermes Agent | OpenClaw |
 |---|---|---|---|---|---|---|
 | 协议注入 | Hook 动态读取 | SessionStart Hook | sessionStart Hook | Plugin + instructions | Skill / `.hermes.md` | Skill |
-| Hook 数量 | 6 | 3 | 3 | Plugin | 0 | 0 |
+| Hook 数量 | 9 | 3 | 3 | Plugin | 0 | 0 |
 | 失败追踪 | 有 | — | — | — | — | — |
 | 斜杠命令 | `/inner-os` | — | — | `inner-os` tool | `/inner-os` | `/inner-os` |
 | 人设切换 | `/inner-os persona` 命令 | 动态（Hook 读取） | 动态（Hook 读取） | Plugin tool | 脚本注入 | 脚本注入 |