auriga-cli 1.30.2 → 1.31.0

package/README.md

@@ -6,6 +6,12 @@ A modular Claude Code harness — install only the parts you need.
 
 This repo itself is a fully configured harness project. You can clone it to see the full setup, or use the CLI to install individual modules into your own project.
 
+Auriga's harness design is inspired by several open-source skill and agent-workflow projects:
+
+- [obra/superpowers skills](https://github.com/obra/superpowers/tree/main/skills)
+- [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills)
+- [mattpocock/skills](https://github.com/mattpocock/skills/tree/main)
+
 ## What's Included
 
 | Module | Description |
@@ -17,23 +23,21 @@ This repo itself is a fully configured harness project. You can clone it to see
 
 ## Quick Start
 
-### Ask your Agent to install
-
-The easiest path is to let your current Agent read the install guide and follow it:
-
-> Run `npx -y auriga-cli guide`, read the guide, then install the Auriga harness into this repository by following the steps it prints.
+### Install
 
-The guide command is intentionally non-interactive. It gives the Agent the prerequisite checks, catalog inspection commands, install commands, reload step, and verification checklist in one place.
+There are two recommended ways to install:
 
-### Agent Bootstrap (non-TTY)
-
-Running inside `claude -p`, `claude -p --worktree`, or any non-interactive Agent session? Start here:
+1. Run the installer yourself:
 
 ```bash
-npx -y auriga-cli guide
+npx -y auriga-cli
 ```
 
-This prints a 5-step SOP (prerequisite check → `install --preset` → optional recommended skills → session reload → verify). Follow it top-to-bottom and the Agent can install the full harness without any human prompt.
+2. In an interactive Agent session, ask the Agent:
+
+> Run `npx -y auriga-cli guide`, read the guide, then install the Auriga harness into this repository by following the steps it prints.
+
+The guide command is intentionally non-interactive so an Agent can read the prerequisite checks, catalog inspection commands, install commands, reload step, and verification checklist in one place.
 
 The leading `-y` belongs to `npx` (it auto-confirms package installation), **not** to `auriga-cli`.
 
@@ -52,7 +56,7 @@ npx -y auriga-cli --help                     # full catalog + flags
 
 `--preset` is atomic — it cannot be combined with a `<type>` or any filter flag, but it accepts `--scope`, `--agent`, and `--lang` (preset defaults: `user` / `both` / `zh-CN`, which differ from the per-category defaults).
 
-Exit codes: `0` success, `1` fatal (precheck / parse / fetch), `2` partial success — `stderr` lists per-category `[OK]/[FAIL]` and a `Retry:` hint. After install, reload the Claude Code or Codex session so the new `AGENTS.md` / skills / plugins / hook-plugin registrations are picked up.
+Exit codes: `0` success, `1` fatal (precheck / parse / fetch), `2` partial success — `stderr` lists per-category `[OK]/[FAIL]` and a `Retry:` hint. After install, reload the Agent session so the new `AGENTS.md` / skills / plugins / hook-plugin registrations are picked up.
 
 ### Web UI (opt-in)
 
@@ -136,7 +140,7 @@ npx -y auriga-cli install plugins --agent codex --plugin session-instructions-lo
 | claude-md-management | Claude Code / Codex | Audit and improve AGENTS.md / CLAUDE.md |
 | playground | Claude Code / Codex | Build interactive HTML playgrounds |
 | codex | Claude Code | Codex cross-model collaboration |
-| auriga-workflow | Claude Code / Codex | The auriga workflow plugin — workflow skills plus the git lifecycle hooks that enforce them. Skills: `incremental-impl`, `test-designer`, `spec-design`, `arch-design`, `code-simplify`, `session-compound`, `goalify` (plans an autonomous goal and dispatches it via the built-in `/goal` command), `deep-review` (multi-dimensional PR review orchestrator — parallel per-dimension reviewers synthesized into an actionable punch list), `reviewer-creator` (scaffolds project-level custom reviewers under `docs/rules/review/`), `git-workflow` (git lifecycle skill), and `documentation-and-adrs` (architecture decision records and project documentation discipline — ADRs archived under `docs/architecture/`). Hooks: `commit-reminder` (PostToolUse on file edits — `Edit` / `Write` / `MultiEdit` in Claude Code, `apply_patch` in Codex — nudges to commit at the next semantic boundary when uncommitted diff vs `HEAD` exceeds 200 lines or 8 files), `pr-create-guard` (PostToolUse on `gh pr create` → injects a PR-body snapshot for five-element self-verification and flags non-Conventional-Commits titles), `pr-ready-guard` (PreToolUse on `gh pr ready` and non-draft `gh pr create` → blocks on stray planning docs, unfinalized active specs under `docs/specs/`, or unpushed commits), and `pr-merge-guard` (PreToolUse on `gh pr merge` → blocks while the PR body's Acceptance criteria or Test plan section still has unchecked checklist items). The two PostToolUse hooks reach full Claude Code / Codex parity; Codex currently fails open on `pr-ready-guard`'s PreToolUse `additionalContext` informational path (block path identical). Installed by default through the plugin path. |
+| auriga-workflow | Claude Code / Codex | The auriga workflow plugin — workflow skills plus the git lifecycle hooks that enforce them. Skills: `incremental-impl`, `test-designer`, `spec-design`, `arch-design`, `code-simplify`, `session-compound`, `goalify` (turns a clarified objective into structured `/goal` text and starts it directly when the runtime supports that), `deep-review` (multi-dimensional PR review orchestrator — parallel per-dimension reviewers synthesized into an actionable punch list), `reviewer-creator` (scaffolds project-level custom reviewers under `docs/rules/review/` at the git repo root), `git-workflow` (git lifecycle skill), and `documentation-and-adrs` (architecture decision records and project documentation discipline — ADRs archived under `docs/architecture/`). Hooks: `commit-reminder` (PostToolUse on file edits — `Edit` / `Write` / `MultiEdit` in Claude Code, `apply_patch` in Codex — nudges to commit at the next semantic boundary when uncommitted diff vs `HEAD` exceeds 200 lines or 8 files), `pr-create-guard` (PostToolUse on `gh pr create` → injects a PR-body snapshot for five-element self-verification and flags non-Conventional-Commits titles), `pr-ready-guard` (PreToolUse on `gh pr ready` and non-draft `gh pr create` → blocks on stray planning docs, unfinalized active specs under `docs/specs/`, or unpushed commits), and `pr-merge-guard` (PreToolUse on `gh pr merge` → blocks while the PR body's Acceptance criteria or Test plan section still has unchecked checklist items). The two PostToolUse hooks reach full Claude Code / Codex parity; Codex currently fails open on `pr-ready-guard`'s PreToolUse `additionalContext` informational path (block path identical). Installed by default through the plugin path. |
 | auriga-notify *(opt-in)* | Claude Code | macOS native notification plugin for Claude Code `Notification` events. Focus-aware sound-only mode, click-to-activate, per-project notification grouping, and migrated `config.json` / `icon.png` support. Not installed by `install --all`; install explicitly with `install plugins --plugin auriga-notify`. |
 | session-instructions-loader | Codex | Codex-only SessionStart plugin that injects ancestor `AGENTS.md` files plus repo-configured extra instruction files. |
 

package/README.zh-CN.md

@@ -6,6 +6,12 @@
 
 这个仓库本身就是一个完整配置好的 harness 项目。可以直接 clone 查看完整配置，也可以用 CLI 把各模块安装到你自己的项目中。
 
+Auriga 的 harness 设计受以下几个开源 skill 与 Agent 工作流项目启发：
+
+- [obra/superpowers skills](https://github.com/obra/superpowers/tree/main/skills)
+- [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills)
+- [mattpocock/skills](https://github.com/mattpocock/skills/tree/main)
+
 ## 包含什么
 
 | 模块 | 说明 |
@@ -17,23 +23,21 @@
 
 ## 快速开始
 
-### 让你的 Agent 负责安装
-
-最简单的方式是让当前 Agent 先读取安装指南，再按指南执行：
-
-> 运行 `npx -y auriga-cli guide`，阅读指南，然后按输出步骤把 Auriga harness 安装到当前仓库。
+### 安装
 
-`guide` 命令是非交互式的。它会把前置检查、catalog 查看命令、安装命令、重启会话步骤和验证清单一次性提供给 Agent。
+推荐安装方式有两种：
 
-### Agent Bootstrap（非交互）
-
-在 `claude -p`、`claude -p --worktree` 或任何非交互 Agent 会话里想装整套 harness？从这里开始：
+1. 用户自己在命令行运行安装器：
 
 ```bash
-npx -y auriga-cli guide
+npx -y auriga-cli
 ```
 
-会打印一份 5 步 SOP（前置检查 → `install --preset` → 可选 recommended skills → 重启 session → 验证）。Agent 照着顺序往下跑就能装完整套 harness，全程不需要人按键。
+2. 在交互式 Agent 会话里输入提示词：
+
+> 运行 `npx -y auriga-cli guide`，阅读指南，然后按输出步骤把 Auriga harness 安装到当前仓库。
+
+`guide` 命令是非交互式的，方便 Agent 一次性读取前置检查、catalog 查看命令、安装命令、重启会话步骤和验证清单。
 
 开头的 `-y` 是 **npx 自己的 flag**（用来跳过"是否要装这个包"的确认），**不是** auriga-cli 的参数。
 
@@ -52,7 +56,7 @@ npx -y auriga-cli --help                     # 完整 catalog + flag 说明
 
 `--preset` 是原子标志 —— 不能与 `<type>` 或任何过滤标志同时使用,但可带 `--scope`、`--agent`、`--lang`(预设默认 `user` / `both` / `zh-CN`,与分类安装的默认不同)。
 
-退出码：`0` 成功；`1` 致命错误（前置检查 / 解析 / 拉取失败）；`2` 部分成功——`stderr` 会列出逐类 `[OK]/[FAIL]` 和 `Retry:` 提示。装完后请重启 Claude Code 或 Codex 会话，让新的 `AGENTS.md` / skills / plugins / hook 插件注册生效。
+退出码：`0` 成功；`1` 致命错误（前置检查 / 解析 / 拉取失败）；`2` 部分成功——`stderr` 会列出逐类 `[OK]/[FAIL]` 和 `Retry:` 提示。装完后请重启 Agent 会话，让新的 `AGENTS.md` / skills / plugins / hook 插件注册生效。
 
 ### Web UI（可选）
 
@@ -136,7 +140,7 @@ npx -y auriga-cli install plugins --agent codex --plugin session-instructions-lo
 | claude-md-management | Claude Code / Codex | 审计和改进 AGENTS.md / CLAUDE.md |
 | playground | Claude Code / Codex | 构建交互式 HTML playground |
 | codex | Claude Code | Codex 跨模型协作 |
-| auriga-workflow | Claude Code / Codex | auriga 工作流插件 —— 工作流 skill 加上强制执行工作流的 git 生命周期 hook。Skills：`incremental-impl`、`test-designer`、`spec-design`、`arch-design`、`code-simplify`、`session-compound`、`goalify`（plan 出自驱 goal 并通过内置 `/goal` 命令分发执行）、`deep-review`（多维度 PR review 编排器——并行派发各维度 reviewer，汇总成可执行的 punch list）、`reviewer-creator`（在 `docs/rules/review/` 下生成项目级自定义 reviewer）、`git-workflow`（git 生命周期 skill）、`documentation-and-adrs`（架构决策记录与项目文档规范，ADR 归档到 `docs/architecture/`）。Hooks：`commit-reminder`（文件编辑的 PostToolUse —— Claude Code 匹配 `Edit` / `Write` / `MultiEdit`，Codex 匹配 `apply_patch` —— 未提交 diff 对比 `HEAD` 超过 200 行或 8 个文件时，提醒在下一个语义边界 commit）、`pr-create-guard`（`gh pr create` 的 PostToolUse —— 注入 PR body 快照供五要素自检，并对不符合 Conventional Commits 的标题提示）、`pr-ready-guard`（`gh pr ready` 与非 draft `gh pr create` 的 PreToolUse —— 拦截游离规划文档、`docs/specs/` 内未结案的活跃 spec、未 push commits）、`pr-merge-guard`（`gh pr merge` 的 PreToolUse —— PR body 的验收标准或测试计划章节仍有未勾选清单项时拦截合并）。两个 PostToolUse hook 在 Claude Code / Codex 上完全对齐；Codex 仅对 `pr-ready-guard` 的 PreToolUse `additionalContext` 信息路径 fail-open（block 路径两边一致）。默认通过插件路径安装。 |
+| auriga-workflow | Claude Code / Codex | auriga 工作流插件 —— 工作流 skill 加上强制执行工作流的 git 生命周期 hook。Skills：`incremental-impl`、`test-designer`、`spec-design`、`arch-design`、`code-simplify`、`session-compound`、`goalify`（把已明确目标整理成结构化 `/goal` 文本，并在运行环境支持时直接启动）、`deep-review`（多维度 PR review 编排器——并行派发各维度 reviewer，汇总成可执行的 punch list）、`reviewer-creator`（在仓库根的 `docs/rules/review/` 下生成项目级自定义 reviewer）、`git-workflow`（git 生命周期 skill）、`documentation-and-adrs`（架构决策记录与项目文档规范，ADR 归档到 `docs/architecture/`）。Hooks：`commit-reminder`（文件编辑的 PostToolUse —— Claude Code 匹配 `Edit` / `Write` / `MultiEdit`，Codex 匹配 `apply_patch` —— 未提交 diff 对比 `HEAD` 超过 200 行或 8 个文件时，提醒在下一个语义边界 commit）、`pr-create-guard`（`gh pr create` 的 PostToolUse —— 注入 PR body 快照供五要素自检，并对不符合 Conventional Commits 的标题提示）、`pr-ready-guard`（`gh pr ready` 与非 draft `gh pr create` 的 PreToolUse —— 拦截游离规划文档、`docs/specs/` 内未结案的活跃 spec、未 push commits）、`pr-merge-guard`（`gh pr merge` 的 PreToolUse —— PR body 的验收标准或测试计划章节仍有未勾选清单项时拦截合并）。两个 PostToolUse hook 在 Claude Code / Codex 上完全对齐；Codex 仅对 `pr-ready-guard` 的 PreToolUse `additionalContext` 信息路径 fail-open（block 路径两边一致）。默认通过插件路径安装。 |
 | auriga-notify *(opt-in)* | Claude Code | Claude Code `Notification` 事件的 macOS 原生通知插件。支持焦点感知仅提示音、点击唤起终端、按项目分组通知，并迁移旧 `config.json` / `icon.png`。不随 `install --all` 默认安装，需要显式执行 `install plugins --plugin auriga-notify`。 |
 | session-instructions-loader | Codex | Codex-only SessionStart 插件，注入上层目录的 `AGENTS.md` 和仓库配置的额外 instruction 文件。 |
 

package/dist/catalog.json

@@ -1,5 +1,5 @@
 {
-  "generatedAt": "2026-05-21T07:29:20.581Z",
+  "generatedAt": "2026-06-10T18:03:54.558Z",
   "workflowSkills": [
     {
       "name": "planning-with-files",
@@ -100,7 +100,7 @@
     },
     {
       "name": "session-instructions-loader",
-      "description": "(Codex) Injects extra instruction files on session start",
+      "description": "(Codex) Injects extra instructions on startup, resume, clear, and compact",
       "agents": [
         "codex"
       ]

package/dist/cli.js

@@ -11,7 +11,7 @@ import { loadCatalog } from "./catalog.js";
 import { renderHelp, renderTypeHelp } from "./help.js";
 import { renderGuide } from "./guide.js";
 import { CATEGORY_NAMES } from "./types.js";
-const RELOAD_REMINDER = "\n⚠ Reload your Claude Code or Codex session to pick up the new harness (AGENTS.md / skills / plugins are loaded at session startup).\n";
+const RELOAD_REMINDER = "\n⚠ Reload your Agent session to pick up the new harness (AGENTS.md / skills / plugins are loaded at session startup).\n";
 const CATEGORY_SET = new Set(CATEGORY_NAMES);
 const TYPE_FOR_FILTER = {
     "--skill": "skills",

package/dist/guide.d.ts

@@ -3,7 +3,7 @@ export interface GuideOpts {
     version: string;
 }
 /**
- * Renders the Agent-bootstrap SOP per spec §3.6. Plain-text when
+ * Renders the interactive Agent install guide. Plain-text when
  * `color: false`; adds ANSI escapes for headings / command examples
  * / warnings when `color: true`. Color detection happens at the call
  * site (`process.stdout.isTTY && !process.env.NO_COLOR`); this

package/dist/guide.js

@@ -7,7 +7,7 @@ function c(color, code, text) {
     return color ? `${code}${text}${RESET}` : text;
 }
 /**
- * Renders the Agent-bootstrap SOP per spec §3.6. Plain-text when
+ * Renders the interactive Agent install guide. Plain-text when
  * `color: false`; adds ANSI escapes for headings / command examples
  * / warnings when `color: true`. Color detection happens at the call
  * site (`process.stdout.isTTY && !process.env.NO_COLOR`); this
@@ -19,9 +19,12 @@ export function renderGuide(opts) {
     const warn = (s) => c(opts.color, YELLOW, s);
     return `${h(`# auriga-cli bootstrap SOP (v${opts.version})`)}
 
-This guide walks an Agent through installing the auriga harness
+This guide helps an interactive Agent install the auriga harness
 (AGENTS.md + skills + plugins) into the current repository.
 
+It is meant to be read by an Agent after the user asks it to install
+Auriga from an interactive session.
+
 Run each step in order. If any step fails with exit 1, stop and report.
 If exit 2, see stderr for per-category status and follow the "Retry"
 hint.
@@ -95,13 +98,12 @@ Exit codes:
   2  — partial success. stderr lists per-category status. Retry only the
        failed category (the retry line is printed verbatim on stderr).
 
-${h("## Step 4 — Reload session (REQUIRED when installed non-interactively)")}
+${h("## Step 4 — Reload session after install")}
 
 ${warn("⚠")} AGENTS.md, .agents/skills/, and plugin enablement /
-registrations are loaded at session startup. If you ran
-\`npx -y auriga-cli install\` inside an existing Claude Code or Codex session
-(e.g., \`claude -p\` / \`claude -p --worktree\` / \`codex exec\`), the current session
-will NOT see the new harness.
+registrations are loaded at session startup. If you installed Auriga
+from an existing Agent session, the current session will NOT see the new
+harness until it is reloaded.
 
 Action:
   - Commit any in-flight work first

package/dist/help.js

@@ -8,7 +8,7 @@ export function renderHelp(catalog, version) {
     return `auriga-cli v${version} — install Claude Code harness modules
 
 USAGE
-  npx auriga-cli guide                                   Agent bootstrap SOP (start here)
+  npx auriga-cli guide                                   interactive Agent install guide
   npx auriga-cli install                                 (TTY only) checkbox menu
   npx auriga-cli install --preset [--scope <s>] [--agent <a>] [--lang <code>]
                                                          curated default set: workflow doc
@@ -24,7 +24,7 @@ USAGE
                                                          open the local Web UI (spec §4)
   npx auriga-cli --help
 
-  For non-interactive (Agent) use, prepend npx's own -y flag:
+  For copy-paste or Agent use, prepend npx's own -y flag:
     npx -y auriga-cli guide
     npx -y auriga-cli install --preset
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "auriga-cli",
-  "version": "1.30.2",
+  "version": "1.31.0",
   "description": "Interactive CLI to install Claude Code harness modules (Workflow, Skills, Recommended Skills, Plugins)",
   "license": "MIT",
   "repository": {
@@ -32,6 +32,7 @@
     "pretest:web-ui-e2e": "npm run build && npm --prefix ui ci && npm --prefix ui run build",
     "test:web-ui-e2e": "tsc -p tsconfig.test.json && node --test dist-test/tests/web-ui-e2e.test.js",
     "test:session-instructions-loader": "node tests/session-instructions-loader.test.mjs",
+    "test:session-compound": "node tests/session-compound-analyzers.test.mjs",
     "test:git-guards": "node tests/commit-reminder.test.mjs && node tests/pr-create-guard.test.mjs && node tests/pr-ready-guard.test.mjs && node tests/pr-merge-guard.test.mjs"
   },
   "engines": {