@zenalexa/unicli 0.224.1 → 0.225.1

package/AGENTS.md

@@ -36,7 +36,7 @@ it starts only the Uni-CLI automation profile.
 
 <!-- BEGIN COUNTS -->
 
-> <!-- STATS:site_count -->312<!-- /STATS --> sites, <!-- STATS:command_count -->1763<!-- /STATS --> commands, <!-- STATS:pipeline_step_count -->103<!-- /STATS --> pipeline steps, BM25 bilingual search. `npm install -g @zenalexa/unicli`
+> <!-- STATS:site_count -->313<!-- /STATS --> sites, <!-- STATS:command_count -->1767<!-- /STATS --> commands, <!-- STATS:pipeline_step_count -->103<!-- /STATS --> pipeline steps, BM25 bilingual search. `npm install -g @zenalexa/unicli`
 
 <!-- END COUNTS -->
 
@@ -46,9 +46,9 @@ it starts only the Uni-CLI automation profile.
 
 ### Web (182+ sites)
 
-**Chinese**: zhihu (27), xiaohongshu (22), bilibili (20), douyin (13), douban (12), v2ex (12), weibo (12), linux-do (11), +29 more (`unicli list`)
+**Chinese**: zhihu (27), xiaohongshu (22), bilibili (20), douyin (13), douban (12), v2ex (12), weibo (12), linux-do (11), +28 more (`unicli list`)
 
-**International**: twitter (47), instagram (29), reddit (24), tiktok (18), youtube (17), nowcoder (16), discord-app (15), facebook (15), +73 more (`unicli list`)
+**International**: twitter (47), instagram (29), reddit (24), tiktok (18), youtube (17), nowcoder (16), discord-app (15), facebook (15), +74 more (`unicli list`)
 
 **AI / ML**: chatgpt (17), antigravity (16), chatwise (16), notebooklm (15), claude (14), doubao-app (13), yollomi (12), deepseek (9), +16 more (`unicli list`)
 
@@ -58,7 +58,7 @@ it starts only the Uni-CLI automation profile.
 
 **News**: hackernews (11), bloomberg (10), 36kr (5), bbc (5), reuters (5), ithome (3), cnn (2), infoq (2), +3 more (`unicli list`)
 
-**Reference**: spotify (23), netease-music (17), linear (10), imdb (7), marxists-cn (7), bitwarden (7), todoist (7), wikipedia (6), +15 more (`unicli list`)
+**Reference**: spotify (24), netease-music (17), linear (10), imdb (7), marxists-cn (7), bitwarden (7), todoist (7), wikipedia (6), +15 more (`unicli list`)
 
 ### macOS (60 cmds)
 
@@ -128,7 +128,7 @@ allowlist entry without a one-line `// REASON:` justification in
 
 ## Version
 
-0.224.1 — Apollo · Collins
+0.225.1 — Apollo · Conrad
 
 ## MCP one-liner (Claude Desktop / Cursor / Continue)
 

package/README.md

@@ -9,20 +9,22 @@
 </p>
 
 <p align="center">
-  <strong>The operations substrate for agents that use real software.</strong>
+  <strong>The universal computer-control platform for agents: an agent control plane for real software.</strong>
 </p>
 
 <p align="center">
-  Uni-CLI turns websites, logged-in browsers, desktop apps, local tools, MCP
-  servers, and system capabilities into searchable, governed, repairable
-  operations. It is not another website wrapper or generated CLI pile; it is the
-  execution layer agents can reuse across runtimes when real software drifts.
+  Install once, search by intent, and run the smallest governed operation across
+  websites, browsers, desktop apps, local tools, files, operating systems, and
+  agent protocols. Browser automation, computer-use sandboxes, MCP servers,
+  local code execution, accessibility, screenshots, app wrappers, and external
+  CLIs are action substrates below one loop: intent in, governed action out,
+  evidence back, repair or reroute until the result is delivered.
 </p>
 
 <p align="center">
   <a href="https://olo-dot-io.github.io/Uni-CLI/">Docs</a>
   ·
-  <a href="https://olo-dot-io.github.io/Uni-CLI/reference/sites">Command catalog</a>
+  <a href="https://olo-dot-io.github.io/Uni-CLI/reference/sites">Operation catalog</a>
   ·
   <a href="https://olo-dot-io.github.io/Uni-CLI/llms.txt">Agent index</a>
 </p>
@@ -39,16 +41,16 @@
 </p>
 
 <p align="center">
-  <sub>Native CLI · MCP · ACP · JSON/Markdown envelopes · browser CDP · visual fallback · macOS desktop AX · <!-- STATS:site_count -->312<!-- /STATS --> surfaces · <!-- STATS:test_count -->8976<!-- /STATS --> tests</sub>
+  <sub>Native CLI · MCP · ACP · JSON/Markdown envelopes · browser CDP · visual fallback · macOS desktop AX · <!-- STATS:site_count -->313<!-- /STATS --> surfaces · <!-- STATS:test_count -->9117<!-- /STATS --> tests</sub>
 </p>
 
 <p align="center">
-  <strong>Search by intent. Execute with policy. Record evidence. Repair the path. Deliver the result.</strong><br>
-  Reuse the same capability from a terminal, an agent loop, MCP, ACP, CI, or a
-  generated skill without rebuilding the integration.
+  <strong>One agent-to-computer control loop. Many substrates. One receipt.</strong><br>
+  Search by intent, act through the best available software boundary, observe
+  the result, and keep the path repairable across CLI, MCP, ACP, CI, and skills.
 </p>
 
-## Start In 30 Seconds
+## Install And Run In 30 Seconds
 
 ```bash
 npm install -g @zenalexa/unicli
@@ -58,67 +60,67 @@ unicli compute snapshot --app Calculator --format compact
 npx @zenalexa/unicli mcp serve
 ```
 
-| Agent problem            | Uni-CLI answer                                                                                     |
-| ------------------------ | -------------------------------------------------------------------------------------------------- |
-| "What tool can do this?" | `unicli search` and `unicli do` turn intent into ranked, inspectable command plans                 |
-| "Can I run it safely?"   | permission profiles expose `open`, `confirm`, and `locked` execution modes                         |
-| "What happened?"         | every run returns an AgentEnvelope with data, context, retryability, and evidence hooks            |
-| "The path failed."       | `unicli delivery` turns run evidence into diagnosis, next experiment, execution, and repair bounds |
-| "The app is local."      | desktop transports cover macOS AX, UIA/AT-SPI sidecars, subprocesses, and visual input             |
-| "Expose it to my agent." | `unicli mcp serve`, ACP, native CLI, and JSON streams share the same catalog                       |
+| Agent problem                 | Uni-CLI answer                                                                                     |
+| ----------------------------- | -------------------------------------------------------------------------------------------------- |
+| "What can control this?"      | `unicli search` and `unicli do` turn intent into ranked operation plans                            |
+| "Which boundary should act?"  | Web, browser, desktop, subprocess, protocol, and visual substrates share one runtime               |
+| "Can I run it safely?"        | permission profiles expose `open`, `confirm`, and `locked` execution modes                         |
+| "What happened?"              | every run returns an AgentEnvelope with data, context, retryability, and evidence hooks            |
+| "The path failed."            | `unicli delivery` turns run evidence into diagnosis, next experiment, execution, and repair bounds |
+| "Expose it to my agent host." | `unicli mcp serve`, ACP, native CLI, and JSON streams expose the same operation contracts          |
 
 ## Why It Exists
 
 The next software user is not only a person with a mouse. It is an agent with a
-task, a context window, a permission budget, and a need for evidence. A raw
-browser driver gives the agent pixels and selectors. A one-off script gives it
-an island. A giant resident tool list burns context before the task starts.
-
-Uni-CLI is the execution layer in between. It turns useful operations into
-typed, searchable commands, keeps dangerous actions behind policy, returns
-machine-readable receipts, and makes failures repairable by pointing at the
-exact adapter and pipeline step that broke. The delivery operator now adds the
-missing model above repair: it classifies why an objective failed, chooses
-retry, route switch, auth, permission, or repair, and keeps the next attempt tied
-to evidence. Websites are just one surface; real agent work crosses browser,
-desktop, subprocess, protocol, and OS boundaries.
-
-This is why the project combines pieces that usually live apart:
-
-- a website adapter catalog for public APIs, logged-in sessions, browser
-  intercepts, downloads, publishing, and search;
-- a browser automation layer for ad-hoc navigation, DOM snapshots, network
-  capture, clicks, typing, screenshots, and render-aware evidence;
-- a local computer-use layer for desktop apps, macOS Accessibility, background
-  input, subprocesses, and visual fallback;
-- a CLI hub for external binaries and agent coding tools;
-- protocol adapters for MCP, ACP, native CLI, JSON streams, and skills.
-
-## The Operating Model
+task, a context window, a permission budget, and a need for evidence. Vehicle
+assistants work because navigation, media, climate, and driving assistance sit
+behind a bounded control surface. General computers need the same idea at larger
+scale: browser state, desktop apps, local tools, files, OS services,
+accessibility trees, screenshots, protocol servers, and website-specific paths
+must become one controllable environment.
+
+Uni-CLI is that control surface. It is not a browser library, a computer-use VM,
+a natural-language shell, an MCP server, or a pile of site wrappers. Those are
+all useful substrates. Uni-CLI sits above them and turns agent intent into
+governed software action with an evidence receipt and a repairable path.
+
+The platform combines pieces that usually live apart, but they are not separate
+product identities:
+
+- operation contracts for reusable actions across websites, apps, OS state,
+  local tools, files, and protocols;
+- substrate adapters for HTTP, browser CDP, desktop accessibility, subprocesses,
+  visual fallback, MCP, ACP, and external CLIs;
+- an invocation kernel that validates args, evaluates policy, executes through
+  the selected substrate, and returns one envelope;
+- an evidence and delivery loop that diagnoses failure, records the next
+  hypothesis, reroutes or repairs, and decides whether the objective is done.
+
+## The Computer-Control Loop
 
 Every serious agent operation follows the same loop.
 
-| Step      | What Uni-CLI gives the agent                                                                                      |
-| --------- | ----------------------------------------------------------------------------------------------------------------- |
-| Discover  | `unicli search` and `unicli do` map intent to command, args, auth mode, examples, and risk signals                |
-| Execute   | Web, browser, desktop, subprocess, and protocol transports run behind one runtime                                 |
-| Govern    | `open`, `confirm`, and `locked` profiles block risky effects before requests, writes, or spawns                   |
-| Observe   | AgentEnvelope v2 returns data, context, retryability, timing, and evidence hooks                                  |
-| Diagnose  | `unicli delivery assess` classifies failures as product drift, missing context, policy block, or upstream trouble |
-| Adapt     | `unicli delivery trajectory` records the hypothesis and next executable experiment                                |
-| Retry     | `unicli delivery run` executes the next safe experiment through the shared kernel and records the new attempt     |
-| Deliver   | Evidence gates decide whether the objective is satisfied, still active, blocked, or exhausted                     |
-| Re-expose | The same command can be called by humans, agents, MCP clients, ACP clients, and scripts                           |
+| Step     | What Uni-CLI gives the agent                                                                                       |
+| -------- | ------------------------------------------------------------------------------------------------------------------ |
+| Intent   | `unicli search` and `unicli do` map a task to candidate operations, args, auth mode, examples, and risk signals    |
+| Select   | operation contracts choose the smallest boundary that can act: API, browser, desktop, subprocess, protocol, visual |
+| Govern   | `open`, `confirm`, and `locked` profiles block risky effects before requests, writes, or spawns                    |
+| Act      | the shared kernel invokes the selected substrate instead of forking wrapper-specific behavior                      |
+| Observe  | AgentEnvelope v2 returns data, context, retryability, timing, and evidence hooks                                   |
+| Diagnose | `unicli delivery assess` classifies failures as product drift, missing context, policy block, or upstream trouble  |
+| Repair   | `unicli delivery trajectory` and `repair-candidate` keep the next experiment bounded by evidence                   |
+| Deliver  | evidence gates decide whether the objective is satisfied, still active, blocked, or exhausted                      |
+| Expose   | the same operation can be called by humans, agents, MCP clients, ACP clients, CI, and scripts                      |
 
 ## What Makes It Different
 
-| If you start from...     | You usually get...                                  | Uni-CLI adds...                                                                 |
-| ------------------------ | --------------------------------------------------- | ------------------------------------------------------------------------------- |
-| Browser automation       | powerful ad-hoc page control                        | catalog commands, output contracts, auth posture, run evidence, and repair      |
-| A generated app CLI      | deep access to one application                      | one searchable runtime across web, desktop, local tools, and agent protocols    |
-| An MCP server collection | easy agent attachment but high resident tool weight | a low-token CLI path plus MCP profiles when the runtime requires MCP            |
-| A scraper script         | quick data extraction                               | policy gates, structured errors, typed args, tests, docs, and local overrides   |
-| Manual computer use      | flexible last-mile control                          | native accessibility first, then browser/visual fallback under the same receipt |
+| If you start from...             | You usually get...                                  | Uni-CLI makes it a substrate under...                                             |
+| -------------------------------- | --------------------------------------------------- | --------------------------------------------------------------------------------- |
+| Browser automation               | powerful page control                               | operation contracts, auth posture, evidence, delivery, and repair                 |
+| Computer-use sandboxes           | screen, mouse, keyboard, and benchmark environments | the same agent-to-computer loop used by local apps, browsers, CLIs, and protocols |
+| Natural-language local execution | flexible shell and code access                      | typed operation boundaries, policy, receipts, and reusable command contracts      |
+| MCP server collections           | easy agent attachment but high resident tool weight | a compact search-first path plus MCP profiles when the host requires MCP          |
+| Per-app or per-site wrappers     | deep access to one surface                          | one governed runtime across web, desktop, local tools, files, and agent protocols |
 
 <!-- IF YOU ARE AN AI AGENT: Start here.
 Install: npm install -g @zenalexa/unicli
@@ -129,24 +131,22 @@ Delivery: write an objective spec, run unicli delivery run spec.json -f json
 Prefer native CLI / JSON stream / MCP for agent runtimes. Use ACP as an editor compatibility gateway.
 -->
 
-## What It Does
-
-Uni-CLI is not another scraper bundle. It is the execution layer underneath an
-agent: a command graph, policy gate, evidence recorder, local computer-use
-bridge, and repair loop wrapped behind one CLI contract.
-
-| Surface            | What you get                                                                                                                                                                                             |
-| ------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| Websites and APIs  | Declarative adapters for public, cookie, header, and browser-intercept workflows                                                                                                                         |
-| Browser automation | CDP steps for navigate, click, type, intercept, snapshot, extract, wait, and related browser work                                                                                                        |
-| Desktop and macOS  | System commands, app adapters, real-time Shortcuts/App Intent discovery, screenshots, clipboard, calendar, brightness, and local tools                                                                   |
-| External CLIs      | 59 registered pass-through bridges with install/status discovery                                                                                                                                         |
-| Agent backends     | Route matrix for native CLI, JSON stream, MCP, ACP, HTTP API, OpenAI-compatible, and bridge routes                                                                                                       |
-| Operation policy   | `open`, `confirm`, and `locked` profiles with effect/risk scopes, local deny rules, `--yes`, and persisted approval memory                                                                               |
-| Evidence           | Run traces with environment snapshots, probe/replay/compare scores, structured gate results, browser session leases with tab/auth posture, render-aware evidence, movement checks, and stale-ref details |
-| Delivery operator  | `unicli delivery assess`, `run`, `trajectory`, and `repair-candidate` for objective-level evidence gates, diagnoses, hypotheses, executed attempts, and bounded repairs                                  |
-| Output             | v2 `AgentEnvelope` in Markdown, JSON, YAML, CSV, or compact format                                                                                                                                       |
-| Repair             | Structured errors with `adapter_path`, failing `step`, retryability, suggestions, and alternatives                                                                                                       |
+## What It Controls
+
+Uni-CLI treats the computer as the environment and each controllable boundary as
+a substrate. The substrate can be high-level and typed, or low-level and visual;
+the receipt remains the same.
+
+| Layer               | What it controls                                                                                                                                                                  |
+| ------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Operation contracts | reusable actions with args, output shape, auth posture, safety metadata, source path, and repair path                                                                             |
+| Web and APIs        | public, cookie, header, browser-intercept, download, upload, publish, extract, and search workflows                                                                               |
+| Browser sessions    | CDP navigate/click/type/fill/select/wait/network/screenshot/snapshot/evidence                                                                                                     |
+| Desktop and OS      | macOS apps, Accessibility refs, screenshots, clipboard, calendar, brightness, app actions, and local system state                                                                 |
+| Local tools/files   | subprocess bridges, external binaries, PDF and paper workflows, file transforms, media tools, and developer CLIs                                                                  |
+| Agent protocols     | native CLI, JSON stream, MCP, ACP, Streamable HTTP, OpenAI-compatible routes, generated configs, and skills                                                                       |
+| Policy and evidence | permission profiles, deny rules, approvals, run recording, replay, probe, compare, browser session leases, render-aware evidence, movement checks, and stale-ref details          |
+| Delivery and repair | `unicli delivery assess`, `run`, `trajectory`, and `repair-candidate` for objective-level evidence gates, diagnoses, hypotheses, executed attempts, reroutes, and bounded repairs |
 
 ## Built For Agent Runtimes
 
@@ -165,19 +165,19 @@ The runtime is agent-friendly in ways that matter during long tasks:
 - repair instructions point to the owned file instead of asking the agent to
   guess what changed upstream;
 - generated docs, `llms.txt`, AGENTS.md, MCP profiles, and skills all describe
-  the same command catalog.
+  the same operation contracts.
 
 ## Capability Map
 
-| Layer                | Examples                                                                                 |
-| -------------------- | ---------------------------------------------------------------------------------------- |
-| Search and discovery | `search`, `do`, generated command catalog, docs index, compact catalog, AGENTS surface   |
-| Web adapters         | HTTP, RSS, cookie, header, browser-intercept, download, upload, publish, extract         |
-| Browser sessions     | CDP open/click/type/fill/select/wait/network/screenshot/snapshot/evidence                |
-| Local computer use   | `compute apps`, `snapshot`, `find`, `click`, `type`, `press`, `scroll`, `doctor compute` |
-| Desktop/system       | macOS, Office, design/media tools, Docker, app actions, subprocess bridges               |
-| Policy and evidence  | permission profiles, deny rules, approvals, run recording, replay, probe, compare        |
-| Integration          | native CLI, MCP stdio, MCP Streamable HTTP, ACP, package exports, agent skills           |
+| Layer                  | Examples                                                                                 |
+| ---------------------- | ---------------------------------------------------------------------------------------- |
+| Intent and discovery   | `search`, `do`, generated operation catalog, docs index, compact catalog, AGENTS surface |
+| Operation contract     | args, output, auth posture, effect, safety, capability, source path, repair path         |
+| Action substrates      | HTTP, RSS, CDP, AX/UIA/AT-SPI, subprocess, visual, protocol, app-specific adapters       |
+| Local computer control | `compute apps`, `snapshot`, `find`, `click`, `type`, `press`, `scroll`, `doctor compute` |
+| Policy and evidence    | permission profiles, deny rules, approvals, run recording, replay, probe, compare        |
+| Delivery and repair    | objective specs, trajectories, repair candidates, reroutes, evidence gates               |
+| Runtime exposure       | native CLI, MCP stdio, MCP Streamable HTTP, ACP, package exports, agent skills           |
 
 ## For Agents
 
@@ -352,7 +352,7 @@ The wall below is generated from active manifest sites with real logo support. B
 <p><strong>audio</strong><br>
   <a data-site="apple-podcasts" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="apple-podcasts: 2 commands"><img alt="apple-podcasts" src="https://img.shields.io/static/v1?label=apple-podcasts&message=2+cmds&color=16a34a&style=flat-square&logo=applepodcasts&logoColor=white"></a>
   <a data-site="netease-music" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="netease-music: 15 commands"><img alt="netease-music" src="https://img.shields.io/static/v1?label=netease-music&message=15+cmds&color=16a34a&style=flat-square&logo=neteasecloudmusic&logoColor=white"></a>
-  <a data-site="spotify" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="spotify: 23 commands"><img alt="spotify" src="https://img.shields.io/static/v1?label=spotify&message=23+cmds&color=16a34a&style=flat-square&logo=spotify&logoColor=white"></a>
+  <a data-site="spotify" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="spotify: 24 commands"><img alt="spotify" src="https://img.shields.io/static/v1?label=spotify&message=24+cmds&color=16a34a&style=flat-square&logo=spotify&logoColor=white"></a>
 </p>
 <p><strong>content</strong><br>
   <a data-site="pixiv" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="pixiv: 6 commands"><img alt="pixiv" src="https://img.shields.io/static/v1?label=pixiv&message=6+cmds&color=c2410c&style=flat-square&logo=pixiv&logoColor=white"></a>
@@ -552,5 +552,5 @@ npm run verify
 [Apache-2.0](./LICENSE)
 
 <p align="center">
-  <sub>v0.224.1 — Apollo · Collins</sub>
+  <sub>v0.225.1 — Apollo · Conrad</sub>
 </p>

package/README.zh-CN.md

@@ -9,18 +9,21 @@
 </p>
 
 <p align="center">
-  <strong>给 AI Agent 操作真实软件的执行底座。</strong>
+  <strong>AI Agent 控制真实软件的通用 computer-control 平台。</strong>
 </p>
 
 <p align="center">
-  Uni-CLI 把网站、登录态浏览器、桌面应用、本地工具、MCP 服务和系统能力收进一套可搜索、可治理、可修复的操作层。
-  它不是又一个网页 wrapper，也不是生成一堆 CLI；它是 Agent 在真实软件持续漂移时仍能复用的执行层。
+  一次安装，按意图搜索，再通过最小可治理操作控制网站、浏览器、桌面 App、
+  本地工具、文件、操作系统和 Agent 协议。浏览器自动化、computer-use sandbox、
+  MCP 服务、本地代码执行、无障碍树、截图、App wrapper 和外部 CLI，
+  都是同一个闭环下面的行动 substrate：意图进入，按策略行动，证据返回，
+  失败后修复或换路，直到结果交付。
 </p>
 
 <p align="center">
   <a href="https://olo-dot-io.github.io/Uni-CLI/">文档站</a>
   ·
-  <a href="https://olo-dot-io.github.io/Uni-CLI/reference/sites">命令目录</a>
+  <a href="https://olo-dot-io.github.io/Uni-CLI/reference/sites">操作目录</a>
   ·
   <a href="https://olo-dot-io.github.io/Uni-CLI/llms.txt">Agent 索引</a>
 </p>
@@ -37,15 +40,15 @@
 </p>
 
 <p align="center">
-  <sub>Native CLI · MCP · ACP · JSON/Markdown envelope · browser CDP · visual fallback · macOS desktop AX · <!-- STATS:site_count -->312<!-- /STATS --> 个 surface · <!-- STATS:test_count -->8976<!-- /STATS --> 个测试</sub>
+  <sub>Native CLI · MCP · ACP · JSON/Markdown envelope · browser CDP · visual fallback · macOS desktop AX · <!-- STATS:site_count -->313<!-- /STATS --> 个 surface · <!-- STATS:test_count -->9117<!-- /STATS --> 个测试</sub>
 </p>
 
 <p align="center">
-  <strong>按意图发现，按策略执行，记录证据，修复路径，交付结果。</strong><br>
-  同一份能力可以从终端、Agent loop、MCP、ACP、CI 或生成的 skill 里调用，不需要为每个入口重写一遍集成。
+  <strong>一条 Agent-to-computer 控制闭环，多种 substrate，同一种回执。</strong><br>
+  按意图搜索，通过最合适的软件边界行动，观察结果，并让路径在 CLI、MCP、ACP、CI 和 skill 中保持可修复。
 </p>
 
-## 30 秒开始
+## 30 秒安装并运行
 
 ```bash
 npm install -g @zenalexa/unicli
@@ -55,54 +58,53 @@ unicli compute snapshot --app Calculator --format compact
 npx @zenalexa/unicli mcp serve
 ```
 
-| Agent 遇到的问题       | Uni-CLI 给的答案                                                                |
-| ---------------------- | ------------------------------------------------------------------------------- |
-| "哪个工具能做这个？"   | `unicli search` 和 `unicli do` 把意图转成可检查、可执行的命令计划               |
-| "能不能安全地跑？"     | permission profile 暴露 `open`、`confirm`、`locked` 三种执行模式                |
-| "刚才发生了什么？"     | 每次运行都返回带 data、context、retryability 和证据钩子的 AgentEnvelope         |
-| "这条路径失败了。"     | `unicli delivery` 把运行证据转成诊断、下一次实验、执行和修复边界                |
-| "目标是本地应用。"     | desktop transport 覆盖 macOS AX、UIA/AT-SPI sidecar、subprocess 和 visual input |
-| "把它接给我的 Agent。" | `unicli mcp serve`、ACP、native CLI、JSON stream 共享同一个目录                 |
+| Agent 遇到的问题            | Uni-CLI 给的答案                                                                 |
+| --------------------------- | -------------------------------------------------------------------------------- |
+| "什么能控制这个？"          | `unicli search` 和 `unicli do` 把意图转成排序后的操作计划                        |
+| "应该走哪个软件边界？"      | Web、browser、desktop、subprocess、protocol、visual substrate 共享同一个 runtime |
+| "能不能安全地跑？"          | permission profile 暴露 `open`、`confirm`、`locked` 三种执行模式                 |
+| "刚才发生了什么？"          | 每次运行都返回带 data、context、retryability 和证据钩子的 AgentEnvelope          |
+| "这条路径失败了。"          | `unicli delivery` 把运行证据转成诊断、下一次实验、执行和修复边界                 |
+| "把它接给我的 Agent host。" | `unicli mcp serve`、ACP、native CLI、JSON stream 暴露同一份操作合同              |
 
 ## 为什么需要它
 
-下一代软件用户不只是拿鼠标的人，也会是带着任务、上下文窗口、权限预算和证据需求的 Agent。直接给浏览器驱动，Agent 得临场猜 selector；写一个脚本，只能解决一个孤岛；把巨大工具列表常驻到上下文里，还没开始任务就先烧掉一截 token。
+下一代软件用户不只是拿鼠标的人，也会是带着任务、上下文窗口、权限预算和证据需求的 Agent。车载助手能成立，是因为导航、媒体、空调、辅助驾驶都在一个有边界的控制层下面。通用 computer 也有同样结构，只是规模更大：浏览器状态、桌面 App、本地工具、文件、OS 服务、无障碍树、截图、协议服务、站点路径，都需要被编译成一个可控环境。
 
-Uni-CLI 做的是中间那层执行底座：把可复用操作整理成 typed、可搜索的命令；把危险动作放到策略闸门后面；把结果变成机器可读回执；失败时明确指出坏掉的 adapter 和 pipeline step。交付 operator 补上 repair 上方缺失的模型：判断一个目标为什么没交付，选择重试、换路径、补认证、请求权限或进入有边界的修复，并让下一次尝试继续绑定证据。网页只是其中一个 surface，真正的 Agent 任务会同时跨过 browser、desktop、subprocess、protocol 和 OS 边界。
+Uni-CLI 就是这个控制面。它不是浏览器库、computer-use VM、自然语言 shell、MCP 服务，也不是一堆站点 wrapper；这些都是有价值的 substrate。Uni-CLI 位于它们之上，把 Agent 的意图变成可治理的软件动作，返回证据回执，并保留可修复、可换路的路径。
 
-所以这个项目把通常分散的几块放在一起：
+因此项目里这些能力不是并列身份，而是同一个平台的下层：
 
-- 网站 adapter catalog：公开 API、登录态会话、浏览器 intercept、下载、发布、搜索；
-- 浏览器自动化层：临场导航、DOM snapshot、网络捕获、点击、输入、截图、render-aware 证据；
-- 本地 computer-use 层：桌面应用、macOS Accessibility、后台输入、subprocess、visual fallback；
-- CLI hub：外部二进制、开发者工具、Agent coding tools；
-- 协议层：MCP、ACP、native CLI、JSON stream、skills。
+- operation contract：覆盖网站、App、OS 状态、本地工具、文件和协议的可复用动作；
+- substrate adapter：HTTP、browser CDP、desktop accessibility、subprocess、visual fallback、MCP、ACP、external CLI；
+- invocation kernel：校验参数、执行策略、选择 substrate、统一 envelope；
+- evidence / delivery loop：诊断失败、记录下一条假设、修复或换路，并判断目标是否已经交付。
 
-## 操作模型
+## Computer-Control 闭环
 
 严肃的 Agent 操作基本都走同一条链路。
 
-| 步骤   | Uni-CLI 给 Agent 的东西                                                               |
-| ------ | ------------------------------------------------------------------------------------- |
-| 发现   | `unicli search` 和 `unicli do` 把一句任务话映射成命令、参数、认证方式、样例和风险信号 |
-| 执行   | Web、browser、desktop、subprocess、protocol transport 走同一个 runtime                |
-| 治理   | `open`、`confirm`、`locked` profile 在请求、写入、启动进程前拦住高风险动作            |
-| 观察   | AgentEnvelope v2 返回 data、context、retryability、耗时和证据 hook                    |
-| 诊断   | `unicli delivery assess` 把失败归为产品漂移、缺少上下文、策略阻断或上游/环境问题      |
-| 调整   | `unicli delivery trajectory` 记录假设和下一次可执行实验                               |
-| 重试   | `unicli delivery run` 通过共享 kernel 执行下一次安全实验，并记录新的 attempt          |
-| 交付   | evidence gate 判断目标已交付、仍在进行、被阻断，还是已经耗尽                          |
-| 再暴露 | 同一条命令可以给人、Agent、MCP client、ACP client 和脚本调用                          |
+| 步骤 | Uni-CLI 给 Agent 的东西                                                                      |
+| ---- | -------------------------------------------------------------------------------------------- |
+| 意图 | `unicli search` 和 `unicli do` 把任务映射成候选操作、参数、认证方式、样例和风险信号          |
+| 选择 | operation contract 选择能行动的最小边界：API、browser、desktop、subprocess、protocol、visual |
+| 治理 | `open`、`confirm`、`locked` profile 在请求、写入、启动进程前拦住高风险动作                   |
+| 行动 | 共享 kernel 调用选中的 substrate，不让 wrapper 各自定义行为                                  |
+| 观察 | AgentEnvelope v2 返回 data、context、retryability、耗时和证据 hook                           |
+| 诊断 | `unicli delivery assess` 把失败归为产品漂移、缺少上下文、策略阻断或上游/环境问题             |
+| 修复 | `unicli delivery trajectory` 和 `repair-candidate` 让下一次实验继续受证据约束                |
+| 交付 | evidence gate 判断目标已交付、仍在进行、被阻断，还是已经耗尽                                 |
+| 暴露 | 同一条操作可以给人、Agent、MCP client、ACP client、CI 和脚本调用                             |
 
 ## 和常见路径的区别
 
-| 如果从...开始       | 通常会得到...                         | Uni-CLI 补上的东西                                                            |
-| ------------------- | ------------------------------------- | ----------------------------------------------------------------------------- |
-| 浏览器自动化        | 强大的临场网页控制                    | 命令目录、输出合同、登录态姿态、运行证据和修复路径                            |
-| 生成某个 App 的 CLI | 对单个应用的深访问                    | 横跨 web、desktop、本地工具和 Agent 协议的一套可搜索 runtime                  |
-| MCP server 集合     | 接 Agent 很方便，但常驻工具上下文很重 | 低 token 的 CLI 主路径，以及运行时需要时才启用的 MCP profile                  |
-| 爬虫脚本            | 快速提取数据                          | 权限策略、结构化错误、typed args、测试、文档和本地 override                   |
-| 手动 computer use   | 灵活的最后一公里控制                  | 优先走 native accessibility，必要时落到 browser/visual fallback，回执仍然统一 |
+| 如果从...开始           | 通常会得到...                     | Uni-CLI 把它放到哪个上位模型下面                                     |
+| ----------------------- | --------------------------------- | -------------------------------------------------------------------- |
+| 浏览器自动化            | 强大的网页控制                    | operation contract、登录态姿态、证据、交付和修复                     |
+| computer-use sandbox    | 屏幕、鼠标、键盘和 benchmark 环境 | 和本地 App、浏览器、CLI、协议共用的 Agent-to-computer 闭环           |
+| 自然语言本地执行        | 灵活的 shell 和代码能力           | typed 操作边界、策略、回执和可复用合同                               |
+| MCP server 集合         | 接 Agent 很方便，但常驻上下文很重 | search-first 的低 token 主路径，以及 host 需要时才启用的 MCP profile |
+| 单 App / 单站点 wrapper | 对一个 surface 的深访问           | 横跨 web、desktop、本地工具、文件和 Agent 协议的一套可治理 runtime   |
 
 <!-- IF YOU ARE AN AI AGENT: Start here.
 Install: npm install -g @zenalexa/unicli
@@ -113,22 +115,20 @@ Delivery: write an objective spec, run unicli delivery run spec.json -f json
 Prefer native CLI / JSON stream / MCP for agent runtimes. Use ACP as an editor compatibility gateway.
 -->
 
-## 它解决什么
+## 它控制什么
 
-Uni-CLI 不是又一个 scraper 包，而是 Agent 底下的执行层：命令图谱、策略闸门、证据记录、本地 computer-use bridge 和自修复 loop，都收在同一个 CLI 合同里。
+Uni-CLI 把 computer 当作环境，把每一个可控边界当作 substrate。substrate 可以是高层 typed API，也可以是低层 visual；回执保持同一种。
 
-| 表面          | 能力                                                                                                                                                                       |
-| ------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| 网站和 API    | public、cookie、header、browser-intercept 等 adapter                                                                                                                       |
-| 浏览器自动化  | CDP 的 navigate、click、type、intercept、snapshot、extract、wait 等步骤                                                                                                    |
-| 桌面和 macOS  | 系统命令、App adapter、实时 Shortcuts/App Intent 发现、截图、剪贴板、日历、亮度、本地工具                                                                                  |
-| 外部 CLI      | 58 个已登记的 passthrough bridge，支持安装和状态发现                                                                                                                       |
-| Agent 后端    | native CLI、JSON stream、MCP、ACP、HTTP API、OpenAI-compatible、bridge 路由矩阵                                                                                            |
-| 操作策略      | `open`、`confirm`、`locked` profile，暴露 effect/risk scope、本地 deny 规则、`--yes` 和持久审批记忆                                                                        |
-| 执行证据      | run trace 会记录环境快照，也能 probe/replay/compare 打分并输出结构化 gate 结果；浏览器 session lease 带 tab/auth 姿态，还支持 render-aware 证据、移动检测和 stale-ref 细节 |
-| 交付 operator | `unicli delivery assess`、`run`、`trajectory`、`repair-candidate`，覆盖 objective-level evidence gate、diagnosis、hypothesis、已执行尝试和有边界修复                       |
-| 输出          | v2 `AgentEnvelope`，支持 Markdown、JSON、YAML、CSV、compact                                                                                                                |
-| 修复          | 错误里带 `adapter_path`、失败 `step`、是否可重试、修复建议和替代命令                                                                                                       |
+| 层级               | 能力                                                                                                                                                         |
+| ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| Operation contract | 可复用动作，带 args、输出形状、认证姿态、安全元数据、source path、repair path                                                                                |
+| Web 和 API         | public、cookie、header、browser-intercept、download、upload、publish、extract、search 工作流                                                                 |
+| Browser session    | CDP navigate/click/type/fill/select/wait/network/screenshot/snapshot/evidence                                                                                |
+| Desktop 和 OS      | macOS App、Accessibility ref、截图、剪贴板、日历、亮度、App action、本地系统状态                                                                             |
+| 本地工具和文件     | subprocess bridge、外部二进制、PDF/论文工作流、文件转换、媒体工具、开发者 CLI                                                                                |
+| Agent 协议         | native CLI、JSON stream、MCP、ACP、Streamable HTTP、OpenAI-compatible route、生成配置、skill                                                                 |
+| 策略和证据         | permission profile、deny rule、approval、run recording、replay、probe、compare、browser session lease、render-aware evidence、movement check、stale-ref 细节 |
+| 交付和修复         | `unicli delivery assess`、`run`、`trajectory`、`repair-candidate`，覆盖 evidence gate、diagnosis、hypothesis、已执行尝试、换路和有边界修复                   |
 
 ## 为 Agent Runtime 设计
 
@@ -140,19 +140,19 @@ Uni-CLI 在边界上刻意保持朴素：进程、文件、JSON、Markdown 和
 - 输出足够稳定，可以 pipe 给 `jq`、存为证据，或者继续喂给下一步工具；
 - auth failure、empty result、timeout、blocked action 是不同 exit state；
 - repair instruction 指向我们拥有的文件，而不是让 Agent 猜 upstream 哪儿变了；
-- 生成文档、`llms.txt`、AGENTS.md、MCP profile 和 skills 描述的是同一份命令目录。
+- 生成文档、`llms.txt`、AGENTS.md、MCP profile 和 skills 描述的是同一份操作合同。
 
 ## 能力地图
 
-| 层级              | 例子                                                                                     |
-| ----------------- | ---------------------------------------------------------------------------------------- |
-| 搜索和发现        | `search`、`do`、生成命令目录、docs index、compact catalog、AGENTS surface                |
-| Web adapter       | HTTP、RSS、cookie、header、browser-intercept、download、upload、publish、extract         |
-| 浏览器 session    | CDP open/click/type/fill/select/wait/network/screenshot/snapshot/evidence                |
-| 本地 computer use | `compute apps`、`snapshot`、`find`、`click`、`type`、`press`、`scroll`、`doctor compute` |
-| 桌面和系统        | macOS、Office、设计/音视频工具、Docker、App actions、subprocess bridge                   |
-| 策略和证据        | permission profile、deny rule、approval、run recording、replay、probe、compare           |
-| 集成              | native CLI、MCP stdio、MCP Streamable HTTP、ACP、package export、agent skills            |
+| 层级                  | 例子                                                                                     |
+| --------------------- | ---------------------------------------------------------------------------------------- |
+| 意图和发现            | `search`、`do`、生成操作目录、docs index、compact catalog、AGENTS surface                |
+| Operation contract    | args、output、auth posture、effect、safety、capability、source path、repair path         |
+| 行动 substrate        | HTTP、RSS、CDP、AX/UIA/AT-SPI、subprocess、visual、protocol、App-specific adapter        |
+| 本地 computer control | `compute apps`、`snapshot`、`find`、`click`、`type`、`press`、`scroll`、`doctor compute` |
+| 策略和证据            | permission profile、deny rule、approval、run recording、replay、probe、compare           |
+| 交付和修复            | objective spec、trajectory、repair candidate、reroute、evidence gate                     |
+| Runtime 暴露          | native CLI、MCP stdio、MCP Streamable HTTP、ACP、package export、agent skills            |
 
 ## 给 Agent 的入口
 
@@ -311,7 +311,7 @@ ACP 作为编辑器和桥接兼容层保留。真正跑任务时，优先 native
 <p><strong>音频</strong><br>
   <a data-site="apple-podcasts" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="apple-podcasts: 2 commands"><img alt="apple-podcasts" src="https://img.shields.io/static/v1?label=apple-podcasts&message=2+cmds&color=16a34a&style=flat-square&logo=applepodcasts&logoColor=white"></a>
   <a data-site="netease-music" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="netease-music: 15 commands"><img alt="netease-music" src="https://img.shields.io/static/v1?label=netease-music&message=15+cmds&color=16a34a&style=flat-square&logo=neteasecloudmusic&logoColor=white"></a>
-  <a data-site="spotify" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="spotify: 23 commands"><img alt="spotify" src="https://img.shields.io/static/v1?label=spotify&message=23+cmds&color=16a34a&style=flat-square&logo=spotify&logoColor=white"></a>
+  <a data-site="spotify" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="spotify: 24 commands"><img alt="spotify" src="https://img.shields.io/static/v1?label=spotify&message=24+cmds&color=16a34a&style=flat-square&logo=spotify&logoColor=white"></a>
 </p>
 <p><strong>内容</strong><br>
   <a data-site="pixiv" href="https://olo-dot-io.github.io/Uni-CLI/reference/sites" title="pixiv: 6 commands"><img alt="pixiv" src="https://img.shields.io/static/v1?label=pixiv&message=6+cmds&color=c2410c&style=flat-square&logo=pixiv&logoColor=white"></a>
@@ -496,5 +496,5 @@ npm run verify
 [Apache-2.0](./LICENSE)
 
 <p align="center">
-  <sub>v0.224.1 — Apollo · Collins</sub>
+  <sub>v0.225.1 — Apollo · Conrad</sub>
 </p>

package/dist/adapters/_electron/desktop-shared.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"desktop-shared.d.ts","sourceRoot":"","sources":["../../../src/adapters/_electron/desktop-shared.ts"],"names":[],"mappings":"AASA,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,6BAA6B;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,oBAAoB,CAAC;CAC9B;AAED,eAAO,MAAM,8BAA8B,iGAQjC,CAAC;AAEX,eAAO,MAAM,+BAA+B,oEAOlC,CAAC;AAOX,wBAAgB,+BAA+B,CAC7C,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,6BAAkC,GAC1C,IAAI,CAyJN"}
+{"version":3,"file":"desktop-shared.d.ts","sourceRoot":"","sources":["../../../src/adapters/_electron/desktop-shared.ts"],"names":[],"mappings":"AAYA,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,6BAA6B;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,KAAK,CAAC,EAAE,oBAAoB,CAAC;CAC9B;AAED,eAAO,MAAM,8BAA8B,iGAQjC,CAAC;AAEX,eAAO,MAAM,+BAA+B,oEAOlC,CAAC;AAQX,wBAAgB,+BAA+B,CAC7C,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,6BAAkC,GAC1C,IAAI,CAyJN"}

package/dist/adapters/_electron/desktop-shared.js

@@ -1,7 +1,7 @@
 import { cli, Strategy } from "../../registry.js";
 import { getElectronApp, resolveAppControlPolicy, } from "../../electron-apps.js";
 import { launchElectronApp } from "../../browser/launcher.js";
-import { connectElectronApp } from "./shared.js";
+import { connectElectronApp, ELECTRON_APP_MINIMUM_CAPABILITY, } from "./shared.js";
 export const ELECTRON_DESKTOP_BASE_COMMANDS = [
     "open-app",
     "status-app",
@@ -22,6 +22,7 @@ export const ELECTRON_DESKTOP_MEDIA_COMMANDS = [
 const ELECTRON_DESKTOP_COMMAND_META = {
     adapter_path: "src/adapters/electron-desktop/electron-desktop.ts",
     target_surface: "desktop",
+    minimum_capability: ELECTRON_APP_MINIMUM_CAPABILITY,
 };
 export function registerElectronDesktopCommands(site, profile = {}) {
     const app = getElectronApp(site);