@opengsd/gsd-pi 1.2.0-dev.5457a158 → 1.2.0-dev.84c56d87

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (135) hide show
  1. package/dist/resources/.managed-resources-content-hash +1 -1
  2. package/dist/resources/extensions/browser-tools/engine/managed-gsd-browser.js +209 -88
  3. package/dist/resources/extensions/browser-tools/engine/selection.js +73 -5
  4. package/dist/resources/extensions/browser-tools/index.js +69 -12
  5. package/dist/resources/extensions/claude-code-cli/stream-adapter.js +3 -2
  6. package/dist/resources/extensions/gsd/bootstrap/register-extension.js +19 -0
  7. package/dist/resources/extensions/gsd/bootstrap/register-hooks.js +27 -9
  8. package/dist/resources/extensions/gsd/browser-evidence.js +8 -2
  9. package/dist/resources/extensions/gsd/mcp-filter.js +2 -19
  10. package/dist/resources/extensions/gsd/uat-policy.js +2 -1
  11. package/dist/resources/extensions/gsd/unit-registry.js +7 -20
  12. package/dist/resources/extensions/gsd/web-app-uat.js +45 -8
  13. package/dist/resources/extensions/search-the-web/native-search.js +5 -3
  14. package/dist/resources/extensions/shared/browser-contract.js +59 -0
  15. package/dist/resources/extensions/shared/gsd-browser-cli.js +72 -4
  16. package/dist/resources/skills/create-skill/references/executable-code.md +1 -1
  17. package/dist/resources/skills/create-skill/workflows/add-reference.md +8 -3
  18. package/dist/resources/skills/create-skill/workflows/add-script.md +4 -2
  19. package/dist/resources/skills/create-skill/workflows/add-template.md +3 -1
  20. package/dist/resources/skills/create-skill/workflows/add-workflow.md +8 -3
  21. package/dist/resources/skills/create-skill/workflows/upgrade-to-router.md +10 -5
  22. package/dist/resources/skills/create-skill/workflows/verify-skill.md +9 -4
  23. package/dist/resources/skills/spike-wrap-up/SKILL.md +9 -9
  24. package/dist/tsconfig.extensions.tsbuildinfo +1 -1
  25. package/dist/web/standalone/.next/BUILD_ID +1 -1
  26. package/dist/web/standalone/.next/app-path-routes-manifest.json +9 -9
  27. package/dist/web/standalone/.next/build-manifest.json +2 -2
  28. package/dist/web/standalone/.next/prerender-manifest.json +3 -3
  29. package/dist/web/standalone/.next/server/app/_global-error.html +1 -1
  30. package/dist/web/standalone/.next/server/app/_global-error.rsc +1 -1
  31. package/dist/web/standalone/.next/server/app/_global-error.segments/_full.segment.rsc +1 -1
  32. package/dist/web/standalone/.next/server/app/_global-error.segments/_global-error/__PAGE__.segment.rsc +1 -1
  33. package/dist/web/standalone/.next/server/app/_global-error.segments/_global-error.segment.rsc +1 -1
  34. package/dist/web/standalone/.next/server/app/_global-error.segments/_head.segment.rsc +1 -1
  35. package/dist/web/standalone/.next/server/app/_global-error.segments/_index.segment.rsc +1 -1
  36. package/dist/web/standalone/.next/server/app/_global-error.segments/_tree.segment.rsc +1 -1
  37. package/dist/web/standalone/.next/server/app/_not-found.html +1 -1
  38. package/dist/web/standalone/.next/server/app/_not-found.rsc +1 -1
  39. package/dist/web/standalone/.next/server/app/_not-found.segments/_full.segment.rsc +1 -1
  40. package/dist/web/standalone/.next/server/app/_not-found.segments/_head.segment.rsc +1 -1
  41. package/dist/web/standalone/.next/server/app/_not-found.segments/_index.segment.rsc +1 -1
  42. package/dist/web/standalone/.next/server/app/_not-found.segments/_not-found/__PAGE__.segment.rsc +1 -1
  43. package/dist/web/standalone/.next/server/app/_not-found.segments/_not-found.segment.rsc +1 -1
  44. package/dist/web/standalone/.next/server/app/_not-found.segments/_tree.segment.rsc +1 -1
  45. package/dist/web/standalone/.next/server/app/index.html +1 -1
  46. package/dist/web/standalone/.next/server/app/index.rsc +1 -1
  47. package/dist/web/standalone/.next/server/app/index.segments/__PAGE__.segment.rsc +1 -1
  48. package/dist/web/standalone/.next/server/app/index.segments/_full.segment.rsc +1 -1
  49. package/dist/web/standalone/.next/server/app/index.segments/_head.segment.rsc +1 -1
  50. package/dist/web/standalone/.next/server/app/index.segments/_index.segment.rsc +1 -1
  51. package/dist/web/standalone/.next/server/app/index.segments/_tree.segment.rsc +1 -1
  52. package/dist/web/standalone/.next/server/app-paths-manifest.json +9 -9
  53. package/dist/web/standalone/.next/server/chunks/8357.js +1 -1
  54. package/dist/web/standalone/.next/server/middleware-build-manifest.js +1 -1
  55. package/dist/web/standalone/.next/server/pages/404.html +1 -1
  56. package/dist/web/standalone/.next/server/pages/500.html +1 -1
  57. package/dist/web/standalone/.next/server/server-reference-manifest.json +1 -1
  58. package/dist/web/standalone/node_modules/node-pty/build/Makefile +1 -1
  59. package/dist/web/standalone/node_modules/postcss/lib/container.js +18 -26
  60. package/dist/web/standalone/node_modules/postcss/lib/css-syntax-error.js +14 -47
  61. package/dist/web/standalone/node_modules/postcss/lib/declaration.js +4 -4
  62. package/dist/web/standalone/node_modules/postcss/lib/fromJSON.js +3 -3
  63. package/dist/web/standalone/node_modules/postcss/lib/input.js +29 -54
  64. package/dist/web/standalone/node_modules/postcss/lib/lazy-result.js +37 -47
  65. package/dist/web/standalone/node_modules/postcss/lib/map-generator.js +9 -26
  66. package/dist/web/standalone/node_modules/postcss/lib/no-work-result.js +55 -57
  67. package/dist/web/standalone/node_modules/postcss/lib/node.js +31 -99
  68. package/dist/web/standalone/node_modules/postcss/lib/parse.js +1 -1
  69. package/dist/web/standalone/node_modules/postcss/lib/parser.js +9 -10
  70. package/dist/web/standalone/node_modules/postcss/lib/postcss.js +12 -12
  71. package/dist/web/standalone/node_modules/postcss/lib/previous-map.js +11 -30
  72. package/dist/web/standalone/node_modules/postcss/lib/processor.js +7 -7
  73. package/dist/web/standalone/node_modules/postcss/lib/result.js +5 -5
  74. package/dist/web/standalone/node_modules/postcss/lib/rule.js +6 -6
  75. package/dist/web/standalone/node_modules/postcss/lib/stringifier.js +28 -69
  76. package/dist/web/standalone/node_modules/postcss/lib/tokenize.js +2 -6
  77. package/dist/web/standalone/node_modules/postcss/package.json +48 -48
  78. package/dist/web/standalone/package.json +1 -1
  79. package/package.json +1 -1
  80. package/packages/cloud-mcp-gateway/package.json +2 -2
  81. package/packages/contracts/package.json +1 -1
  82. package/packages/daemon/package.json +4 -4
  83. package/packages/gsd-agent-core/package.json +5 -5
  84. package/packages/gsd-agent-modes/package.json +7 -7
  85. package/packages/mcp-server/package.json +3 -3
  86. package/packages/native/package.json +1 -1
  87. package/packages/pi-agent-core/package.json +1 -1
  88. package/packages/pi-ai/dist/models.generated.d.ts +66 -178
  89. package/packages/pi-ai/dist/models.generated.d.ts.map +1 -1
  90. package/packages/pi-ai/dist/models.generated.js +116 -204
  91. package/packages/pi-ai/dist/models.generated.js.map +1 -1
  92. package/packages/pi-ai/package.json +1 -1
  93. package/packages/pi-coding-agent/package.json +7 -7
  94. package/packages/pi-tui/dist/tui.d.ts.map +1 -1
  95. package/packages/pi-tui/dist/tui.js +9 -0
  96. package/packages/pi-tui/dist/tui.js.map +1 -1
  97. package/packages/pi-tui/package.json +2 -2
  98. package/packages/rpc-client/package.json +2 -2
  99. package/pkg/package.json +1 -1
  100. package/src/resources/extensions/browser-tools/engine/managed-gsd-browser.ts +265 -98
  101. package/src/resources/extensions/browser-tools/engine/selection.ts +90 -4
  102. package/src/resources/extensions/browser-tools/index.ts +71 -13
  103. package/src/resources/extensions/browser-tools/tests/browser-engine-selection.test.mjs +83 -13
  104. package/src/resources/extensions/browser-tools/tests/managed-gsd-browser-tools.test.mjs +136 -0
  105. package/src/resources/extensions/claude-code-cli/stream-adapter.ts +3 -2
  106. package/src/resources/extensions/gsd/bootstrap/register-extension.ts +24 -0
  107. package/src/resources/extensions/gsd/bootstrap/register-hooks.ts +48 -4
  108. package/src/resources/extensions/gsd/browser-evidence.ts +18 -2
  109. package/src/resources/extensions/gsd/mcp-filter.ts +2 -23
  110. package/src/resources/extensions/gsd/tests/browser-automation-contract-fixture.ts +39 -0
  111. package/src/resources/extensions/gsd/tests/browser-contract.test.ts +44 -0
  112. package/src/resources/extensions/gsd/tests/dispatch-run-uat-browser-tools.test.ts +2 -1
  113. package/src/resources/extensions/gsd/tests/extension-bootstrap-isolation.test.ts +35 -1
  114. package/src/resources/extensions/gsd/tests/integration/auto-worktree-milestone-merge.test.ts +7 -11
  115. package/src/resources/extensions/gsd/tests/integration/auto-worktree.test.ts +20 -58
  116. package/src/resources/extensions/gsd/tests/integration/gsd-integration-fixture.ts +80 -0
  117. package/src/resources/extensions/gsd/tests/token-tool-gating.test.ts +35 -0
  118. package/src/resources/extensions/gsd/tests/uat-policy.test.ts +24 -29
  119. package/src/resources/extensions/gsd/tests/web-app-uat.test.ts +44 -1
  120. package/src/resources/extensions/gsd/uat-policy.ts +2 -1
  121. package/src/resources/extensions/gsd/unit-registry.ts +7 -20
  122. package/src/resources/extensions/gsd/web-app-uat.ts +51 -8
  123. package/src/resources/extensions/search-the-web/native-search.ts +5 -3
  124. package/src/resources/extensions/shared/browser-contract.ts +66 -0
  125. package/src/resources/extensions/shared/gsd-browser-cli.ts +88 -4
  126. package/src/resources/skills/create-skill/references/executable-code.md +1 -1
  127. package/src/resources/skills/create-skill/workflows/add-reference.md +8 -3
  128. package/src/resources/skills/create-skill/workflows/add-script.md +4 -2
  129. package/src/resources/skills/create-skill/workflows/add-template.md +3 -1
  130. package/src/resources/skills/create-skill/workflows/add-workflow.md +8 -3
  131. package/src/resources/skills/create-skill/workflows/upgrade-to-router.md +10 -5
  132. package/src/resources/skills/create-skill/workflows/verify-skill.md +9 -4
  133. package/src/resources/skills/spike-wrap-up/SKILL.md +9 -9
  134. /package/dist/web/standalone/.next/static/{2p9Rv9pQflAxCBbGVI2vb → AOpDeK_gJHU8OZjRo31gQ}/_buildManifest.js +0 -0
  135. /package/dist/web/standalone/.next/static/{2p9Rv9pQflAxCBbGVI2vb → AOpDeK_gJHU8OZjRo31gQ}/_ssgManifest.js +0 -0
@@ -1,11 +1,38 @@
1
+ /**
2
+ * Browser Automation Engine resolution (ADR-037).
3
+ *
4
+ * The engine choice behind the canonical `browser_*` tools is a runtime
5
+ * decision with a recorded reason, not a static default:
6
+ * - An explicit `GSD_BROWSER_ENGINE` override is honored verbatim.
7
+ * - Otherwise, browser-facing projects prefer the managed gsd-browser engine
8
+ * when the availability probe can prove a CLI exists, and fall back to
9
+ * legacy Playwright with the failure reason recorded.
10
+ * - Non-browser-facing projects keep legacy Playwright (browser tools are
11
+ * incidental there; the managed daemon is not worth its startup risk).
12
+ *
13
+ * This module owns the committed resolution, not just the prediction: when
14
+ * registration verifies a probe-resolved managed engine (daemon connect) and
15
+ * falls back, the outcome is committed here so every ambient reader — UAT
16
+ * guidance, re-warm-up, later sessions in the same process — sees the engine
17
+ * the session actually registered.
18
+ */
19
+ import path from "node:path";
20
+
21
+ import { resolveGsdBrowserCliAvailability } from "../../shared/gsd-browser-cli.js";
22
+ import { detectWebApp } from "../web-app-detect.js";
23
+
1
24
  export type BrowserEngineMode = "gsd-browser" | "legacy" | "off";
2
25
 
3
- const DEFAULT_BROWSER_ENGINE: BrowserEngineMode = "legacy";
26
+ export interface BrowserEngineResolution {
27
+ engine: BrowserEngineMode;
28
+ /** "env" = explicit GSD_BROWSER_ENGINE override; "probe" = default path decided by availability. */
29
+ source: "env" | "probe";
30
+ reason: string;
31
+ }
4
32
 
5
- export function resolveBrowserEngineMode(env: NodeJS.ProcessEnv = process.env): BrowserEngineMode {
6
- const raw = env.GSD_BROWSER_ENGINE?.trim();
7
- if (!raw) return DEFAULT_BROWSER_ENGINE;
33
+ const committedResolutionByProjectRoot = new Map<string, BrowserEngineResolution>();
8
34
 
35
+ function parseExplicitEngineMode(raw: string): BrowserEngineMode {
9
36
  const normalized = raw.toLowerCase();
10
37
  if (normalized === "gsd-browser" || normalized === "gsd_browser" || normalized === "gsdbrowser") {
11
38
  return "gsd-browser";
@@ -17,3 +44,62 @@ export function resolveBrowserEngineMode(env: NodeJS.ProcessEnv = process.env):
17
44
 
18
45
  throw new Error(`Invalid GSD_BROWSER_ENGINE="${raw}". Expected "gsd-browser", "legacy", or "off".`);
19
46
  }
47
+
48
+ /** Pure resolution from explicit inputs. Never cached; probes on every call. */
49
+ export function resolveBrowserEngineResolution(
50
+ env: NodeJS.ProcessEnv,
51
+ projectRoot?: string,
52
+ ): BrowserEngineResolution {
53
+ const raw = env.GSD_BROWSER_ENGINE?.trim();
54
+ if (raw) {
55
+ return { engine: parseExplicitEngineMode(raw), source: "env", reason: `GSD_BROWSER_ENGINE=${raw}` };
56
+ }
57
+
58
+ if (!projectRoot) {
59
+ return { engine: "legacy", source: "probe", reason: "no project root to probe; using legacy Playwright" };
60
+ }
61
+
62
+ if (!detectWebApp(projectRoot)) {
63
+ return {
64
+ engine: "legacy",
65
+ source: "probe",
66
+ reason: "project is not browser-facing; using legacy Playwright",
67
+ };
68
+ }
69
+
70
+ const availability = resolveGsdBrowserCliAvailability(env);
71
+ return availability.available
72
+ ? {
73
+ engine: "gsd-browser",
74
+ source: "probe",
75
+ reason: `web app detected and managed gsd-browser engine available (${availability.detail})`,
76
+ }
77
+ : {
78
+ engine: "legacy",
79
+ source: "probe",
80
+ reason: `web app detected but gsd-browser unavailable (${availability.detail}); falling back to legacy Playwright`,
81
+ };
82
+ }
83
+
84
+ /**
85
+ * Session-facing resolution: the committed record for this project root, or
86
+ * the ambient probe result, cached as the initial commitment (the probe
87
+ * touches the filesystem and at worst one short subprocess).
88
+ */
89
+ export function resolveAmbientBrowserEngineResolution(projectRoot: string): BrowserEngineResolution {
90
+ const key = path.resolve(projectRoot);
91
+ const committed = committedResolutionByProjectRoot.get(key);
92
+ if (committed) return committed;
93
+
94
+ const resolution = resolveBrowserEngineResolution(process.env, projectRoot);
95
+ committedResolutionByProjectRoot.set(key, resolution);
96
+ return resolution;
97
+ }
98
+
99
+ /**
100
+ * Record a verified outcome for this project root — e.g. the probe predicted
101
+ * gsd-browser but the daemon-connect gate fell back to legacy Playwright.
102
+ */
103
+ export function commitBrowserEngineResolution(projectRoot: string, resolution: BrowserEngineResolution): void {
104
+ committedResolutionByProjectRoot.set(path.resolve(projectRoot), resolution);
105
+ }
@@ -2,7 +2,7 @@
2
2
  import { importExtensionModule, type ExtensionAPI, type ExtensionContext } from "@gsd/pi-coding-agent";
3
3
 
4
4
  import { closeManagedGsdBrowser, registerManagedGsdBrowserTools, warmUpManagedGsdBrowser } from "./engine/managed-gsd-browser.js";
5
- import { resolveBrowserEngineMode, type BrowserEngineMode } from "./engine/selection.js";
5
+ import { commitBrowserEngineResolution, resolveAmbientBrowserEngineResolution, type BrowserEngineMode } from "./engine/selection.js";
6
6
  import { setArtifactRootForCwd } from "./state.js";
7
7
  import { detectWebApp } from "./web-app-detect.js";
8
8
 
@@ -167,13 +167,59 @@ function withBrowserArtifactCwdScope(pi: ExtensionAPI): ExtensionAPI {
167
167
  };
168
168
  }
169
169
 
170
- async function registerBrowserTools(pi: ExtensionAPI): Promise<void> {
171
- const engine = resolveBrowserEngineMode();
170
+ /** Daemon-connect budget when the probe-resolved managed engine is verified at session start. */
171
+ const PROBE_WARMUP_TIMEOUT_MS = 10_000;
172
+
173
+ async function registerBrowserTools(pi: ExtensionAPI, ctx: ExtensionContext): Promise<void> {
174
+ const projectRoot = ctx.cwd || process.cwd();
175
+ const resolution = resolveAmbientBrowserEngineResolution(projectRoot);
176
+ let engine = resolution.engine;
172
177
  if (engine === "off") return;
178
+
179
+ // A probe-resolved managed engine is only a prediction that gsd-browser
180
+ // works — prove it by connecting the daemon before committing the session's
181
+ // tool registrations to it. Connect failure falls back to legacy Playwright
182
+ // (the failure mode that made ADR-024 freeze the old default) and commits
183
+ // the outcome so ambient readers see the engine actually in use. When eager
184
+ // warm-up is disabled the daemon-connect proof cannot run, so the probe
185
+ // default treats the managed engine as unprovable and falls back to legacy
186
+ // rather than registering it unverified. An explicit
187
+ // GSD_BROWSER_ENGINE=gsd-browser override skips the gate and is honored
188
+ // verbatim, matching prior behavior.
189
+ if (engine === "gsd-browser" && resolution.source === "probe" && !registeredEngine) {
190
+ if (isWarmUpDisabled()) {
191
+ engine = commitLegacyFallback(projectRoot, "warm-up disabled; managed engine unverifiable; using legacy Playwright");
192
+ } else {
193
+ const warmUp = await warmUpManagedGsdBrowser(ctx, AbortSignal.timeout(PROBE_WARMUP_TIMEOUT_MS));
194
+ if (!warmUp.ok) {
195
+ engine = commitLegacyFallback(projectRoot, `gsd-browser daemon connect failed (${warmUp.error}); using legacy Playwright`);
196
+ if (ctx.hasUI) {
197
+ ctx.ui.notify(
198
+ `gsd-browser engine unavailable (${warmUp.error}); using Playwright browser tools for this session.`,
199
+ "warning",
200
+ );
201
+ }
202
+ } else if (warmUp.coverageWarning && ctx.hasUI) {
203
+ ctx.ui.notify(warmUp.coverageWarning, "warning");
204
+ }
205
+ }
206
+ }
207
+
208
+ // Browser tool registrations are process-global and cannot be swapped once
209
+ // live. When an earlier session in this process already registered an engine
210
+ // and this project resolved a different one (per-project probe resolution can
211
+ // diverge across projects in a multi-session process), adopt the registered
212
+ // engine rather than throwing — a throw surfaces as "browser-tools failed to
213
+ // load" and leaves this session with no browser tools at all. Commit the
214
+ // adoption so ambient readers (UAT guidance, warm-up) describe the engine
215
+ // actually in use.
173
216
  if (registeredEngine && registeredEngine !== engine) {
174
- throw new Error(
175
- `Browser tools already registered with GSD_BROWSER_ENGINE=${registeredEngine}. Restart GSD before switching to ${engine}.`,
176
- );
217
+ engine = registeredEngine;
218
+ commitBrowserEngineResolution(projectRoot, {
219
+ engine,
220
+ source: "probe",
221
+ reason: `browser tools already registered with ${engine} earlier in this process; adopting it`,
222
+ });
177
223
  }
178
224
 
179
225
  let registration: Promise<void>;
@@ -202,29 +248,41 @@ async function registerBrowserTools(pi: ExtensionAPI): Promise<void> {
202
248
  }
203
249
  }
204
250
 
251
+ function commitLegacyFallback(projectRoot: string, reason: string): "legacy" {
252
+ commitBrowserEngineResolution(projectRoot, { engine: "legacy", source: "probe", reason });
253
+ return "legacy";
254
+ }
255
+
205
256
  function isWarmUpDisabled(): boolean {
206
257
  const value = process.env.GSD_BROWSER_WARMUP?.trim().toLowerCase();
207
258
  return value === "0" || value === "false" || value === "off";
208
259
  }
209
260
 
210
261
  /**
211
- * Auto-initialize the managed gsd-browser engine only when explicitly selected
212
- * for a web app. Best-effort and non-blocking: warm-up runs in the background
213
- * and only surfaces a warning if it fails.
262
+ * Auto-initialize the managed gsd-browser engine when it was selected via the
263
+ * explicit GSD_BROWSER_ENGINE override, which registers without the
264
+ * daemon-connect gate. Best-effort and non-blocking: warm-up runs in the
265
+ * background and only surfaces a warning if it fails. Probe-resolved sessions
266
+ * already connected (or fell back) during registration, so they are excluded
267
+ * to avoid re-warming and double-notifying.
214
268
  */
215
269
  function maybeWarmUpManagedEngine(pi: ExtensionAPI, ctx: ExtensionContext): void {
216
270
  if (isWarmUpDisabled()) return;
217
- if (resolveBrowserEngineMode() !== "gsd-browser") return;
218
271
 
219
272
  const projectRoot = ctx.cwd || process.cwd();
273
+ const resolution = resolveAmbientBrowserEngineResolution(projectRoot);
274
+ if (resolution.engine !== "gsd-browser" || resolution.source !== "env") return;
220
275
  if (!detectWebApp(projectRoot)) return;
221
276
 
222
277
  void warmUpManagedGsdBrowser(ctx).then((result) => {
223
- if (!result.ok && ctx.hasUI) {
278
+ if (!ctx.hasUI) return;
279
+ if (!result.ok) {
224
280
  ctx.ui.notify(
225
281
  `gsd-browser auto-init failed: ${result.error}. Browser UAT tools will retry on first use; run /gsd doctor if this persists.`,
226
282
  "warning",
227
283
  );
284
+ } else if (result.coverageWarning) {
285
+ ctx.ui.notify(result.coverageWarning, "warning");
228
286
  }
229
287
  });
230
288
  }
@@ -240,7 +298,7 @@ async function closeActiveBrowserEngines(): Promise<void> {
240
298
  export default function (pi: ExtensionAPI) {
241
299
  pi.on("session_start", async (_event, ctx) => {
242
300
  if (ctx.hasUI) {
243
- void registerBrowserTools(pi)
301
+ void registerBrowserTools(pi, ctx)
244
302
  .then(() => maybeWarmUpManagedEngine(pi, ctx))
245
303
  .catch((error) => {
246
304
  ctx.ui.notify(`browser-tools failed to load: ${error instanceof Error ? error.message : String(error)}`, "warning");
@@ -248,7 +306,7 @@ export default function (pi: ExtensionAPI) {
248
306
  return;
249
307
  }
250
308
 
251
- await registerBrowserTools(pi);
309
+ await registerBrowserTools(pi, ctx);
252
310
  maybeWarmUpManagedEngine(pi, ctx);
253
311
  });
254
312
 
@@ -1,35 +1,105 @@
1
1
  import { describe, it } from "node:test";
2
2
  import assert from "node:assert/strict";
3
+ import { mkdtempSync, writeFileSync } from "node:fs";
3
4
  import { createRequire } from "node:module";
4
- import { dirname } from "node:path";
5
+ import { tmpdir } from "node:os";
6
+ import { dirname, join } from "node:path";
5
7
  import { fileURLToPath } from "node:url";
6
8
 
7
9
  const __dirname = dirname(fileURLToPath(import.meta.url));
8
10
  const require = createRequire(import.meta.url);
9
11
  const jiti = require("jiti")(__dirname, { interopDefault: true, debug: false });
10
12
 
11
- const { resolveBrowserEngineMode } = jiti("../engine/selection.ts");
13
+ const {
14
+ commitBrowserEngineResolution,
15
+ resolveAmbientBrowserEngineResolution,
16
+ resolveBrowserEngineResolution,
17
+ } = jiti("../engine/selection.ts");
12
18
 
13
- describe("resolveBrowserEngineMode", () => {
14
- it("defaults to the Playwright engine", () => {
15
- assert.equal(resolveBrowserEngineMode({}), "legacy");
16
- });
19
+ function makeProject({ webApp }) {
20
+ const dir = mkdtempSync(join(tmpdir(), "gsd-engine-selection-"));
21
+ const pkg = webApp ? { dependencies: { react: "^18.0.0" } } : { name: "cli-tool" };
22
+ writeFileSync(join(dir, "package.json"), JSON.stringify(pkg));
23
+ return dir;
24
+ }
25
+
26
+ function makeFakeCli() {
27
+ const dir = mkdtempSync(join(tmpdir(), "gsd-fake-cli-"));
28
+ const cliPath = join(dir, "gsd-browser");
29
+ writeFileSync(cliPath, "#!/bin/sh\n");
30
+ return cliPath;
31
+ }
17
32
 
18
- it("accepts the explicit engine modes", () => {
19
- assert.equal(resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "gsd-browser" }), "gsd-browser");
20
- assert.equal(resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "legacy" }), "legacy");
21
- assert.equal(resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "off" }), "off");
33
+ describe("resolveBrowserEngineResolution", () => {
34
+ it("honors the explicit engine modes verbatim with env source", () => {
35
+ assert.deepEqual(
36
+ ["gsd-browser", "legacy", "off"].map(
37
+ (mode) => resolveBrowserEngineResolution({ GSD_BROWSER_ENGINE: mode }).engine,
38
+ ),
39
+ ["gsd-browser", "legacy", "off"],
40
+ );
41
+ assert.equal(resolveBrowserEngineResolution({ GSD_BROWSER_ENGINE: "gsd-browser" }).source, "env");
22
42
  });
23
43
 
24
44
  it("accepts compatibility aliases", () => {
25
- assert.equal(resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "playwright" }), "legacy");
26
- assert.equal(resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "false" }), "off");
45
+ assert.equal(resolveBrowserEngineResolution({ GSD_BROWSER_ENGINE: "playwright" }).engine, "legacy");
46
+ assert.equal(resolveBrowserEngineResolution({ GSD_BROWSER_ENGINE: "false" }).engine, "off");
27
47
  });
28
48
 
29
49
  it("rejects unknown engine modes", () => {
30
50
  assert.throws(
31
- () => resolveBrowserEngineMode({ GSD_BROWSER_ENGINE: "surprise" }),
51
+ () => resolveBrowserEngineResolution({ GSD_BROWSER_ENGINE: "surprise" }),
32
52
  /Expected "gsd-browser", "legacy", or "off"/,
33
53
  );
34
54
  });
55
+
56
+ it("defaults to legacy Playwright when no project root is known", () => {
57
+ const resolution = resolveBrowserEngineResolution({});
58
+ assert.equal(resolution.engine, "legacy");
59
+ assert.equal(resolution.source, "probe");
60
+ });
61
+
62
+ it("keeps legacy Playwright for non-browser-facing projects", () => {
63
+ const cliPath = makeFakeCli();
64
+ const resolution = resolveBrowserEngineResolution(
65
+ { GSD_BROWSER_CLI_PATH: cliPath },
66
+ makeProject({ webApp: false }),
67
+ );
68
+ assert.equal(resolution.engine, "legacy");
69
+ assert.match(resolution.reason, /not browser-facing/);
70
+ });
71
+
72
+ it("prefers the managed gsd-browser engine for web apps when the CLI is provable", () => {
73
+ const cliPath = makeFakeCli();
74
+ const resolution = resolveBrowserEngineResolution(
75
+ { GSD_BROWSER_CLI_PATH: cliPath },
76
+ makeProject({ webApp: true }),
77
+ );
78
+ assert.equal(resolution.engine, "gsd-browser");
79
+ assert.equal(resolution.source, "probe");
80
+ assert.match(resolution.reason, /web app detected/);
81
+ });
82
+
83
+ it("falls back to legacy Playwright with a recorded reason when the CLI is unavailable", () => {
84
+ const resolution = resolveBrowserEngineResolution(
85
+ { GSD_BROWSER_CLI_PATH: "/nonexistent/gsd-browser" },
86
+ makeProject({ webApp: true }),
87
+ );
88
+ assert.equal(resolution.engine, "legacy");
89
+ assert.equal(resolution.source, "probe");
90
+ assert.match(resolution.reason, /falling back to legacy Playwright/);
91
+ });
92
+ });
93
+
94
+ describe("committed resolution", () => {
95
+ it("ambient readers see a committed verification outcome instead of the prediction", () => {
96
+ const projectRoot = makeProject({ webApp: true });
97
+ const fallback = {
98
+ engine: "legacy",
99
+ source: "probe",
100
+ reason: "gsd-browser daemon connect failed (test); using legacy Playwright",
101
+ };
102
+ commitBrowserEngineResolution(projectRoot, fallback);
103
+ assert.deepEqual(resolveAmbientBrowserEngineResolution(projectRoot), fallback);
104
+ });
35
105
  });
@@ -2,10 +2,33 @@ import { describe, it } from "node:test";
2
2
  import assert from "node:assert/strict";
3
3
 
4
4
  const {
5
+ MANAGED_BROWSER_TOOL_SPECS,
5
6
  MANAGED_GSD_BROWSER_TOOL_NAMES,
7
+ findMissingContractCoverage,
8
+ normalizeManagedArgs,
6
9
  registerManagedGsdBrowserTools,
7
10
  } = await import("../engine/managed-gsd-browser.ts");
8
11
 
12
+ // The tools @opengsd/gsd-browser actually serves over MCP (subset relevant to
13
+ // the contract). Notably absent: browser_click, browser_type, browser_verify,
14
+ // browser_reload — those are satisfied through translations.
15
+ const GSD_BROWSER_SERVED_TOOLS = [
16
+ "browser_navigate",
17
+ "browser_snapshot",
18
+ "browser_click_ref",
19
+ "browser_fill_ref",
20
+ "browser_fill_form",
21
+ "browser_wait_for",
22
+ "browser_assert",
23
+ "browser_screenshot",
24
+ "browser_find_element",
25
+ "browser_console",
26
+ "browser_network",
27
+ "browser_evaluate",
28
+ "browser_batch",
29
+ "browser_act",
30
+ ];
31
+
9
32
  describe("registerManagedGsdBrowserTools", () => {
10
33
  it("registers the curated Pi browser contract", () => {
11
34
  const tools = [];
@@ -31,3 +54,116 @@ describe("registerManagedGsdBrowserTools", () => {
31
54
  assert.equal(screenshot?.compatibility?.producesImages, true);
32
55
  });
33
56
  });
57
+
58
+ describe("findMissingContractCoverage", () => {
59
+ it("reports nothing for the tool list gsd-browser actually serves", () => {
60
+ assert.deepEqual(findMissingContractCoverage(GSD_BROWSER_SERVED_TOOLS), []);
61
+ });
62
+
63
+ it("reports contract tools none of whose MCP candidates are served", () => {
64
+ const served = GSD_BROWSER_SERVED_TOOLS.filter((name) => name !== "browser_assert");
65
+ // browser_verify also depends on browser_assert through its translation.
66
+ assert.deepEqual(findMissingContractCoverage(served), ["browser_assert", "browser_verify"]);
67
+ });
68
+
69
+ it("reports translated tools when a required MCP tool is missing", () => {
70
+ const served = GSD_BROWSER_SERVED_TOOLS.filter((name) => name !== "browser_batch");
71
+ assert.deepEqual(findMissingContractCoverage(served), ["browser_click", "browser_type", "browser_batch"]);
72
+ });
73
+ });
74
+
75
+ describe("contract tool translations", () => {
76
+ it("translates browser_click into a single-step batch call", () => {
77
+ const calls = MANAGED_BROWSER_TOOL_SPECS.browser_click.translate.build({ selector: "#save" });
78
+ assert.deepEqual(calls, [{
79
+ mcpTool: "browser_batch",
80
+ args: { steps: [{ action: "click", selector: "#save" }] },
81
+ }]);
82
+ });
83
+
84
+ it("translates browser_type into a single-step batch call", () => {
85
+ const calls = MANAGED_BROWSER_TOOL_SPECS.browser_type.translate.build({
86
+ selector: "#name",
87
+ text: "hello",
88
+ clearFirst: true,
89
+ submit: true,
90
+ });
91
+ assert.deepEqual(calls, [{
92
+ mcpTool: "browser_batch",
93
+ args: { steps: [{ action: "type", selector: "#name", text: "hello", clearFirst: true, submit: true }] },
94
+ }]);
95
+ });
96
+
97
+ it("normalizes batch options and step keys to the daemon's snake_case", () => {
98
+ const normalized = normalizeManagedArgs("browser_batch", {
99
+ steps: [{ action: "type", selector: "#name", text: "hi", clearFirst: true }],
100
+ stopOnFailure: false,
101
+ finalSummaryOnly: true,
102
+ });
103
+ assert.deepEqual(normalized, {
104
+ steps: [{ action: "type", selector: "#name", text: "hi", clear_first: true }],
105
+ stop_on_failure: false,
106
+ summary_only: true,
107
+ });
108
+ });
109
+
110
+ it("translates browser_verify into navigate, assert, and screenshot calls", () => {
111
+ const calls = MANAGED_BROWSER_TOOL_SPECS.browser_verify.translate.build({
112
+ url: "http://localhost:3000",
113
+ timeout: 5000,
114
+ checks: [
115
+ { description: "heading shows", selector: "h1", expectedText: "Welcome" },
116
+ { description: "spinner gone", selector: ".spinner", expectedVisible: false },
117
+ { description: "evidence", selector: "main", expectedVisible: true, screenshot: true },
118
+ ],
119
+ });
120
+ assert.deepEqual(calls, [
121
+ { mcpTool: "browser_navigate", args: { url: "http://localhost:3000", timeout: 5000 } },
122
+ {
123
+ mcpTool: "browser_assert",
124
+ args: {
125
+ checks: [
126
+ { kind: "text_visible", text: "Welcome" },
127
+ { kind: "selector_hidden", selector: ".spinner" },
128
+ { kind: "selector_visible", selector: "main" },
129
+ ],
130
+ },
131
+ },
132
+ { mcpTool: "browser_screenshot", args: {}, optional: true },
133
+ ]);
134
+ });
135
+
136
+ it("declares every tool a translation can emit in its coverage requirements", () => {
137
+ for (const [name, spec] of Object.entries(MANAGED_BROWSER_TOOL_SPECS)) {
138
+ if (!spec.translate) continue;
139
+ const maximalArgs = {
140
+ url: "http://localhost:3000",
141
+ timeout: 5000,
142
+ selector: "#el",
143
+ text: "hi",
144
+ clearFirst: true,
145
+ checks: [{ description: "d", selector: "#el", expectedText: "hi", expectedVisible: true, screenshot: true }],
146
+ };
147
+ const emitted = spec.translate.build(maximalArgs).map((call) => call.mcpTool);
148
+ for (const mcpTool of emitted) {
149
+ assert.ok(
150
+ spec.translate.requires.includes(mcpTool),
151
+ `${name} translation emits ${mcpTool} but does not require it for coverage`,
152
+ );
153
+ }
154
+ }
155
+ });
156
+
157
+ it("translates browser_verify without checks into navigation only", () => {
158
+ const calls = MANAGED_BROWSER_TOOL_SPECS.browser_verify.translate.build({ url: "http://localhost:3000", checks: [] });
159
+ assert.deepEqual(calls, [{ mcpTool: "browser_navigate", args: { url: "http://localhost:3000" } }]);
160
+ });
161
+
162
+ it("translates browser_reload into evaluate plus best-effort network-idle wait", () => {
163
+ const calls = MANAGED_BROWSER_TOOL_SPECS.browser_reload.translate.build({});
164
+ assert.deepEqual(calls, [
165
+ { mcpTool: "browser_evaluate", args: { expression: "location.reload()" } },
166
+ { mcpTool: "browser_wait_for", args: { condition: "network_idle", timeout: 3_000 }, optional: true },
167
+ ]);
168
+ });
169
+ });
@@ -60,6 +60,7 @@ import {
60
60
  } from "../gsd/mcp-filter.js";
61
61
  import { RUN_UAT_CLAUDE_NATIVE_TOOL_NAMES, RUN_UAT_FORBIDDEN_TOOL_NAMES, RUN_UAT_WORKFLOW_TOOL_NAMES, resolveToolPresentationPlan } from "../gsd/tool-presentation-plan.js";
62
62
  import { getToolSurfaceReadinessError } from "../gsd/tool-surface-readiness.js";
63
+ import { hasBrowserContractPrefix } from "../shared/browser-contract.js";
63
64
  import { showInterviewRound, type Question, type RoundResult } from "../shared/tui.js";
64
65
  import type {
65
66
  SDKAssistantMessage,
@@ -1591,7 +1592,7 @@ function browserMcpServerNameFromAllowedTools(allowedTools: unknown): string | u
1591
1592
  if (typeof toolName !== "string") continue;
1592
1593
  const parsed = parseAllowedMcpToolName(toolName);
1593
1594
  if (!parsed) continue;
1594
- if (parsed.server === "gsd-browser" || parsed.tool.startsWith("browser_")) {
1595
+ if (parsed.server === "gsd-browser" || hasBrowserContractPrefix(parsed.tool)) {
1595
1596
  return parsed.server;
1596
1597
  }
1597
1598
  }
@@ -1604,7 +1605,7 @@ function workflowMcpServerNameFromAllowedTools(allowedTools: unknown): string |
1604
1605
  for (const toolName of allowedTools) {
1605
1606
  if (typeof toolName !== "string") continue;
1606
1607
  const parsed = parseAllowedMcpToolName(toolName);
1607
- if (!parsed || parsed.server === browserServerName || parsed.tool.startsWith("browser_")) continue;
1608
+ if (!parsed || parsed.server === browserServerName || hasBrowserContractPrefix(parsed.tool)) continue;
1608
1609
  return parsed.server;
1609
1610
  }
1610
1611
  return undefined;
@@ -17,6 +17,7 @@ import { registerHooks } from "./register-hooks.js";
17
17
  import { registerShortcuts } from "./register-shortcuts.js";
18
18
  import { writeCrashLog } from "./crash-log.js";
19
19
  import { logWarning } from "../workflow-logger.js";
20
+ import { UNIT_TOOL_CONTRACTS } from "../unit-tool-contracts.js";
20
21
  // Static import so cmux event listeners are registered synchronously during
21
22
  // extension bootstrap. Prior implementation used `void import().then()` which
22
23
  // queued listener registration as a microtask — any CMUX_CHANNELS emit fired
@@ -37,6 +38,12 @@ const EPIPE_STORM_WINDOW_MS = 10_000;
37
38
  let epipeCount = 0;
38
39
  let epipeWindowStart = 0;
39
40
 
41
+ export const CRITICAL_GSD_WORKFLOW_TOOL_NAMES = [...new Set(
42
+ Object.values(UNIT_TOOL_CONTRACTS)
43
+ .flatMap((contract) => contract.requiredWorkflowTools)
44
+ .filter((toolName) => toolName.startsWith("gsd_")),
45
+ )].sort();
46
+
40
47
  /** Write to stderr without ever re-throwing — stderr can EPIPE too, which would
41
48
  * re-enter this handler and re-loop. */
42
49
  function safeStderr(msg: string): void {
@@ -133,6 +140,21 @@ export function installEpipeGuard(): void {
133
140
  }
134
141
  }
135
142
 
143
+ function assertCriticalGsdWorkflowToolsRegistered(pi: ExtensionAPI): void {
144
+ if (typeof pi.getAllTools !== "function") return;
145
+
146
+ const registered = new Set(pi.getAllTools().map((tool) => tool.name));
147
+ const missing = CRITICAL_GSD_WORKFLOW_TOOL_NAMES.filter((toolName) => !registered.has(toolName));
148
+ if (missing.length === 0) return;
149
+
150
+ const message = [
151
+ `Critical GSD workflow tool registration failed; missing required tool(s): ${missing.join(", ")}.`,
152
+ "Check earlier bootstrap warnings for the registration slot that failed.",
153
+ ].join(" ");
154
+ logWarning("bootstrap", message);
155
+ throw new Error(message);
156
+ }
157
+
136
158
  export function registerGsdExtension(pi: ExtensionAPI): void {
137
159
  // Note: registerGSDCommand is called by index.ts before this function,
138
160
  // so we intentionally skip it here to avoid double-registration.
@@ -214,4 +236,6 @@ export function registerGsdExtension(pi: ExtensionAPI): void {
214
236
  );
215
237
  }
216
238
  }
239
+
240
+ assertCriticalGsdWorkflowToolsRegistered(pi);
217
241
  }