fix(dsh)!: namespace BrowserSkill tools and recover failed activation

AGENT_INSTALL.md

@@ -131,7 +131,7 @@ For CLI harnesses, start with `bsk session start --no-focus --json`, retain the
 returned session ID, then navigate and observe using `--session <id>`. Stop that
 session with `bsk session stop <id>` on success or failure. With multiple browsers,
 use `bsk browsers` and add `--browser <id>` when starting the session.
-For dsh, use its injected `browser_*` tools instead.
+For dsh, use its injected `bsk_*` tools instead.
 
 For the WorkBuddy/CodeBuddy or child-reaping-host workflow, use separate tool
 calls for startup, status, session creation, navigation/observation and final

CHANGELOG.md

@@ -7,8 +7,22 @@ Starting from 0.2.0, CLI / Extension / DSH Plugin share the same version number.
 
 ## [Unreleased]
 
+### Changed
+
+- **Breaking — DSH Plugin:** replace `browser_session`, `browser_page`,
+  `browser_inspect`, `browser_interact`, `browser_tabs`, and `browser_assist` with
+  the corresponding `bsk_*` names to coexist with other browser plugins.
+  Actions, arguments, results, and the `browser-skill` skill name are unchanged.
+  Update hardcoded calls and tool policies, restart the host, and reload the skill
+  in existing conversations. See the [migration guide](packages/dsh-plugin-browserskill/README.md#breaking-tool-api-migration-unreleased).
+
 ### Fixed
 
+- DSH Plugin: roll back incomplete tool registration, surface activation failures,
+  and retry at skill/session lifecycle boundaries without leaving a partial suite.
+  Include tool schemas in the first `/browser-skill` request, and keep failure and
+  recovery notices synchronized with model skill results
+  ([#444](https://github.com/Tencent/BrowserSkill/issues/444)).
 - Extension: input to a background Agent Window tab no longer keeps failing with
   `input_not_ready` after Chrome drops the session's focus override without a detach
   ([#355](https://github.com/Tencent/BrowserSkill/issues/355)). The session's

README.md

@@ -154,7 +154,7 @@ In WorkBuddy/CodeBuddy, or hosts that reap command children, the agent should re
 
 ## DeepSeek Harness plugin
 
-The [DSH plugin](packages/dsh-plugin-browserskill/README.md) adds native `browser_*` tools, browser task previews, and screenshot results to the DeepSeek Harness Web UI. It uses the same `bsk` CLI and extension and includes its own BrowserSkill skill.
+The [DSH plugin](packages/dsh-plugin-browserskill/README.md) adds native `bsk_*` tools, browser task previews, and screenshot results to the DeepSeek Harness Web UI. It uses the same `bsk` CLI and extension and includes its own BrowserSkill skill.
 
 With DeepSeek Harness, pnpm, and `bsk` installed and the extension connected, add the plugin to your profile:
 

README.zh-CN.md

@@ -154,7 +154,7 @@ bsk session stop <id>
 
 ## DeepSeek Harness 插件
 
-[DSH 插件](packages/dsh-plugin-browserskill/README.md) 在 DeepSeek Harness Web UI 中提供原生 `browser_*` 工具、浏览器任务预览和截图结果。它使用同一套 `bsk` CLI 与扩展，并自带 BrowserSkill skill。
+[DSH 插件](packages/dsh-plugin-browserskill/README.md) 在 DeepSeek Harness Web UI 中提供原生 `bsk_*` 工具、浏览器任务预览和截图结果。它使用同一套 `bsk` CLI 与扩展，并自带 BrowserSkill skill。
 
 安装好 DeepSeek Harness、pnpm 和 `bsk`，连接扩展后，将插件加入你的 profile：
 

apps/extension/src/entrypoints/popup/profile-instructions.test.tsx

@@ -23,7 +23,7 @@ describe("profile instructions", () => {
     const text = vi.mocked(navigator.clipboard.writeText).mock.calls[0]?.[0];
     expect(text).toContain('browser "Work" (instance ID: a1234567)');
     expect(text).toContain("bsk session start --browser a1234567 --json");
-    expect(text).toContain('browser_session({ action: "start", browser: "a1234567" })');
+    expect(text).toContain('bsk_session({ action: "start", browser: "a1234567" })');
     expect(text).toContain("use the tool call instead of running the CLI command separately");
     expect(text).toContain("every new session for this task");
     expect(text).toContain("Do not omit --browser / browser or switch to another instance");
@@ -70,7 +70,7 @@ describe("profile instructions", () => {
       "bsk session start --browser b1234567 --json",
     );
     expect(vi.mocked(navigator.clipboard.writeText).mock.calls[1]?.[0]).toContain(
-      'browser_session({ action: "start", browser: "b1234567" })',
+      'bsk_session({ action: "start", browser: "b1234567" })',
     );
   });
   it("reports clipboard failure and allows retry without claiming success", async () => {
@@ -90,7 +90,7 @@ describe("profile instructions", () => {
     const text = vi.mocked(navigator.clipboard.writeText).mock.calls[0]?.[0];
     expect(text).toContain(JSON.stringify(label));
     expect(text).toContain("CLI: bsk session start --browser a1234567 --json");
-    expect(text).toContain('browser_session({ action: "start", browser: "a1234567" })');
+    expect(text).toContain('bsk_session({ action: "start", browser: "a1234567" })');
     expect(await screen.findByRole("button", { name: "Copied" })).toBeTruthy();
   });
   it("discards pending feedback when the saved name changes and copies the new name", async () => {
@@ -153,7 +153,7 @@ describe("profile instructions", () => {
       "bsk session start --browser a1234567 --json",
     );
     expect(vi.mocked(navigator.clipboard.writeText).mock.calls[0]?.[0]).toContain(
-      'browser_session({ action: "start", browser: "a1234567" })',
+      'bsk_session({ action: "start", browser: "a1234567" })',
     );
     expect(vi.mocked(navigator.clipboard.writeText).mock.calls[0]?.[0]).toContain(
       "--browser / browser",

apps/extension/src/entrypoints/popup/profile-instructions.tsx

@@ -20,7 +20,7 @@ export function ProfileInstructions({
     name: JSON.stringify(label || t("popup.profile.unnamed")),
     instanceId,
     command: `bsk session start --browser ${instanceId} --json`,
-    toolCall: `browser_session({ action: "start", browser: ${JSON.stringify(instanceId)} })`,
+    toolCall: `bsk_session({ action: "start", browser: ${JSON.stringify(instanceId)} })`,
   });
 
   useEffect(() => {

apps/extension/src/video/capture.browser.test.ts

@@ -730,20 +730,20 @@ describe.skipIf(!process.env.BSK_VIDEO_CHROME || !process.env.BSK_VIDEO_BSK)(
             // navigation and exports after its owned session is gone.
             const dsh = await createDshVideoFixture(executable, env, path.join(directory, "dsh"));
             try {
-              const task = await dsh.call<{ sessionId: string }>("browser_session", {
+              const task = await dsh.call<{ sessionId: string }>("bsk_session", {
                 action: "start",
               });
               const started = await dsh.call<{
                 recording: { recording_id: string; state: string; tab_id: number };
-              }>("browser_inspect", {
+              }>("bsk_inspect", {
                 action: "video",
                 videoAction: "start",
                 session: task.sessionId,
                 durationMs: 30000,
               });
               expect(started.recording.state).toBe("recording");
               const recordingId = started.recording.recording_id;
-              await dsh.call("browser_page", {
+              await dsh.call("bsk_page", {
                 action: "navigate",
                 session: task.sessionId,
                 url: `http://127.0.0.1:${address.port}/red`,
@@ -764,28 +764,28 @@ describe.skipIf(!process.env.BSK_VIDEO_CHROME || !process.env.BSK_VIDEO_BSK)(
               await new Promise((resolve) => setTimeout(resolve, 1200));
               const finished = await dsh.call<{
                 recording: { state: string; completeness: string; byte_size: number };
-              }>("browser_inspect", { action: "video", videoAction: "stop", recordingId });
+              }>("bsk_inspect", { action: "video", videoAction: "stop", recordingId });
               expect(finished.recording.state).toBe("ready");
               expect(finished.recording.completeness).toBe("complete");
-              await dsh.call("browser_session", { action: "stop", session: task.sessionId });
+              await dsh.call("bsk_session", { action: "stop", session: task.sessionId });
               const output = path.join(directory, "dsh.mp4");
-              await dsh.call("browser_inspect", {
+              await dsh.call("bsk_inspect", {
                 action: "video",
                 videoAction: "save",
                 recordingId,
                 output,
               });
               expect((await stat(output)).size).toBe(finished.recording.byte_size);
               await expect(
-                dsh.call("browser_inspect", {
+                dsh.call("bsk_inspect", {
                   action: "video",
                   videoAction: "save",
                   recordingId,
                   output,
                 }),
               ).rejects.toThrow(/Output exists/);
               const list = await dsh.call<{ recordings: { recording_id: string }[] }>(
-                "browser_inspect",
+                "bsk_inspect",
                 { action: "video", videoAction: "list" },
               );
               expect(list.recordings.map((value) => value.recording_id)).toEqual([recordingId]);

docs/browser-profiles.md

@@ -55,7 +55,7 @@ routing identifiers, not independent proof of a filesystem path.
 Use the verified popup instance ID in the plugin's session tool:
 
 ```text
-browser_session({ action: "start", browser: "<verified-instance-id>" })
+bsk_session({ action: "start", browser: "<verified-instance-id>" })
 ```
 
 **Copy instructions** includes this tool call alongside the CLI command.

docs/scroll-to.md

@@ -9,7 +9,7 @@ distance or scroll direction. Use [wheel](wheel.md) for native directional wheel
 | --- | --- |
 | CLI | `bsk scroll-to` |
 | Wire protocol | `tool.scroll_to` |
-| DeepSeek Harness | `browser_interact` with `action: "scroll-to"` |
+| DeepSeek Harness | `bsk_interact` with `action: "scroll-to"` |
 
 ## CLI
 
@@ -97,7 +97,7 @@ The plugin exposes the same action with its usual camelCase parameters:
 }
 ```
 
-Call this through `browser_interact` using a plugin-owned session. Its result
+Call this through `bsk_interact` using a plugin-owned session. Its result
 contains `session`, `tabId`, `x`, `y`, `width` and `height` with the same bounds
 semantics. Omitted `session` uses the plugin's current session.
 

docs/structured-extraction.md

@@ -186,12 +186,12 @@ allows a consumer to detect a mismatched pair.
 
 ## DSH plugin
 
-The existing **browser_inspect** tool gains **action: "extract"**:
+The existing **bsk_inspect** tool gains **action: "extract"**:
 
 ~~~text
-browser_inspect({ action: "extract", session: "<id>", extractKind: "discover" })
-browser_inspect({ action: "extract", session: "<id>", extractKind: "table", extractTarget: "<target_id>" })
-browser_inspect({ action: "extract", session: "<id>", extractKind: "table", selector: "#orders",
+bsk_inspect({ action: "extract", session: "<id>", extractKind: "discover" })
+bsk_inspect({ action: "extract", session: "<id>", extractKind: "table", extractTarget: "<target_id>" })
+bsk_inspect({ action: "extract", session: "<id>", extractKind: "table", selector: "#orders",
   extractFormat: "csv", output: "/absolute/path/orders.csv", csvSafe: true })
 ~~~
 
@@ -203,8 +203,8 @@ requires an output path so the plugin always receives structured JSON. The plugi
 normal session registry, runner and queue; it does not introduce arbitrary
 page-script evaluation. Output paths are on the CLI host; relative paths resolve
 against that process's working directory. Prefer absolute paths in harness calls.
-Extraction shares browser_inspect's output parameter with debug export and video
-save. Its model guidance lives in the browser_inspect description and parameter
+Extraction shares bsk_inspect's output parameter with debug export and video
+save. Its model guidance lives in the bsk_inspect description and parameter
 schema rather than a separate DSH skill reference, keeping the DSH skill entry
 within its 4,500-byte budget in both LF and CRLF checkouts.
 

docs/video-recording.md

@@ -179,7 +179,7 @@ BSK_VIDEO_CHROME=/path/to/chrome BSK_VIDEO_BSK=../../target/debug/bsk \
 ```
 
 The extension's minimum Chrome version remains 125. Codec support is validated at
-runtime. DSH uses `browser_inspect(action=video)` with `videoAction` set to start, status,
+runtime. DSH uses `bsk_inspect(action=video)` with `videoAction` set to start, status,
 stop, save, list, or discard. Its six public tool schemas are retained. Start
 follows the session queue; artifact operations bypass that queue so status/stop
 remain available during human help. A plugin-local allowlist of returned recording
@@ -190,11 +190,11 @@ this allowlist; the extension library remains available for recovery. CLI partia
 stop/save errors retain their recording metadata and saved path in the tool error.
 
 ```text
-browser_inspect({ action: "video", videoAction: "start", session: "<task>", durationMs: 60000 })
+bsk_inspect({ action: "video", videoAction: "start", session: "<task>", durationMs: 60000 })
 // Execute task operations after state=recording.
-browser_inspect({ action: "video", videoAction: "stop", recordingId: "<recording_id>" })
+bsk_inspect({ action: "video", videoAction: "stop", recordingId: "<recording_id>" })
 // End the task. Preview and Save As remain available in the extension.
-browser_inspect({ action: "video", videoAction: "save", recordingId: "<recording_id>", output: "<explicit-path.mp4>" })
+bsk_inspect({ action: "video", videoAction: "save", recordingId: "<recording_id>", output: "<explicit-path.mp4>" })
 ```
 
 

docs/website-debugging.md

@@ -67,7 +67,7 @@ Follow `next_offset` while present. Offsets count UTF-16 code units, matching th
 extension. RFC 6901 `--pointer` works only on a complete retained JSON body and is
 applied after redaction. No network request is repeated by these reads.
 
-DSH uses `browser_inspect` with `action: "debug"`, `debugAction` matching the CLI
+DSH uses `bsk_inspect` with `action: "debug"`, `debugAction` matching the CLI
 subcommand, and camelCase options (`runId`, `tabId`, `maxChars`). The six public
 tools, existing session ownership, cancellation and queue remain unchanged.
 The wire endpoint is `tool.debug`; schemas are in `crates/bsk-protocol/schema`.
@@ -420,7 +420,7 @@ bsk debug replay <request-id> --session <id> --replay-file replay.json
 ```
 
 `--rule` and `--replay` also accept inline JSON. DSH exposes these as JSON-string
-`rule` / `replay` parameters on `browser_inspect(action: "debug")`, with the same
+`rule` / `replay` parameters on `bsk_inspect(action: "debug")`, with the same
 `debugAction` values. File input avoids putting large bodies in command arguments.
 
 Replay sends a **new** request and may change server data. It uses a dedicated

docs/wheel.md

@@ -8,7 +8,7 @@ when the goal is simply to bring an existing element into view.
 | --- | --- |
 | CLI | `bsk wheel` |
 | Wire protocol | `tool.wheel` |
-| DeepSeek Harness | `browser_interact` with `action: "wheel"` |
+| DeepSeek Harness | `bsk_interact` with `action: "wheel"` |
 
 ## Usage