codex-chatgpt-control 0.2.0-alpha.1 → 0.3.0-alpha.1

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.3.0-alpha.1
+
+- Hardens mode-menu detection and selection against thread/sidebar action menus, with locale-registry-backed thread-action vetoes and container-scoped menu enumeration.
+- Adds the `modes.get` primitive and post-selection verification warnings on `modes.set`.
+- Rewrites wait polling around a single combined DOM snapshot per poll; response text is fetched once at completion instead of every poll.
+- Adds Windows and Linux clipboard capture with DOM fallback.
+- Fixes report `createdAt` to honor the injected clock for deterministic fixtures.
+
 ## 0.2.0-alpha.1
 
 - Adds Windows-safe host path validation and cross-platform backend gates.

package/contracts/v1/fixtures/backend-capabilities.json

@@ -45,6 +45,7 @@
     "projects.sources.planAdd",
     "projects.sources.add",
     "modes.set",
+    "modes.get",
     "tools.select",
     "response.copy"
   ],

package/contracts/v1/fixtures/command-descriptors.json

@@ -651,7 +651,13 @@
       "summary": "Compose, submit, optionally wait, and optionally read."
     },
     {
-      "args": {},
+      "args": {
+        "mode": "normal or deep_research",
+        "pollMs": "optional polling interval",
+        "responseContent": "include for current behavior, metadata to omit assistant text and return chars/hash only",
+        "stableMs": "optional stability window before completion",
+        "timeoutMs": "optional wait timeout"
+      },
       "blockers": [
         "browser_bridge_unavailable",
         "login_required",
@@ -779,6 +785,7 @@
     },
     {
       "args": {
+        "includeHashes": "optional SHA-256 metadata for diagnostics; file contents are not returned",
         "maxBytesPerFile": "optional local per-file byte limit",
         "maxTotalBytes": "optional local total byte limit",
         "paths": "absolute local file paths"
@@ -791,7 +798,8 @@
       "defaults": {},
       "defaultTimeoutMs": 30000,
       "examples": [
-        "await chatgpt.files.preflight({ paths: [\"/absolute/host/path.md\"] });"
+        "await chatgpt.files.preflight({ paths: [\"/absolute/host/path.md\"] });",
+        "await chatgpt.files.preflight({ paths: [\"/absolute/host/path.md\"], includeHashes: true });"
       ],
       "layer": "primitive",
       "name": "files.preflight",
@@ -801,6 +809,8 @@
     },
     {
       "args": {
+        "includeDiagnostics": "optional preflight and browser input file-size metadata",
+        "includeHashes": "optional SHA-256 metadata inside diagnostics; file contents are not returned",
         "paths": "absolute local file paths"
       },
       "blockers": [
@@ -959,6 +969,27 @@
       "risk": "medium",
       "summary": "Select a visible model, intelligence, effort, or nested model-version candidate when unambiguous."
     },
+    {
+      "args": {
+        "timeoutMs": "optional timeout for attaching to the ChatGPT page"
+      },
+      "blockers": [
+        "browser_bridge_unavailable",
+        "login_required",
+        "selector_drift"
+      ],
+      "defaults": {},
+      "defaultTimeoutMs": 30000,
+      "examples": [
+        "await chatgpt.modes.get();",
+        "// Verify an expensive Pro consult before submitting: const current = await chatgpt.modes.get();"
+      ],
+      "layer": "primitive",
+      "name": "modes.get",
+      "retryPolicy": "Return structured CommandResult failures; retry only when the blocker is recoverable and no duplicate prompt will be submitted.",
+      "risk": "low",
+      "summary": "Read the mode labels shown on the visible composer controls without changing them."
+    },
     {
       "args": {},
       "blockers": [

package/contracts/v1/fixtures/doctor-scenario-preflight.json

@@ -71,8 +71,19 @@
         "localization": {
           "details": {
             "englishCanonicalPresent": true,
-            "labelCandidateCount": 1070,
+            "labelCandidateCount": 1542,
+            "missingModeOptionIds": [],
             "missingToolIds": [],
+            "modeOptionIds": [
+              "latest",
+              "instant",
+              "thinking",
+              "extended",
+              "medium",
+              "high",
+              "extraHigh",
+              "pro"
+            ],
             "pageAvailable": false,
             "requiredKeysMissing": [],
             "runtimeSelectorCoverage": "registry_only_stage_2",

package/contracts/v1/fixtures/help-root.json

@@ -1,6 +1,6 @@
 {
   "ok": true,
   "requestId": "req_help",
-  "result": "ChatGPT browser-control SDK commands\n\nworkflow:\n- ask: Ask ChatGPT in a new or selected thread, optionally with files, wait/read, downloads, and reports.\n- askInThread: Open or claim an existing thread by URL, conversation id, title, or search query, then ask and read.\n- askWithFiles: Attach absolute local file paths, optionally set mode, ask, wait, and read.\n- askAndDownload: Ask ChatGPT to produce a visible downloadable output and save the latest exposed file.\n- runMessages: Run sequential prompts where later prompts can use earlier step data.\n- runner.run: Agents-style facade: run a visible ChatGPT browser-control agent against input, files, thread, existing-tab, mode, and response options.\n- responses.create: Narrow Responses-shaped adapter over the visible ChatGPT browser-control runner; rejects unsupported API-only fields before prompt submission.\n- copyLatest: Copy or DOM-read the latest assistant response with Markdown-first fidelity.\n- runPlan: Execute an inline SequencePlan or named macro through the existing sequence engine.\n- new-ask-read: Named macro: open a new thread, ask, wait, and read Markdown.\n- find-open-ask-read: Named macro: search history, open the first match, ask, wait, and read Markdown.\n- find-open-copy-latest: Named macro: search history, open the first match, and copy/read the latest response.\n- attach-ask-read: Named macro: open a new thread, attach files, ask, wait, and read Markdown.\n- ask-and-download: Named macro: ask in a new thread and download the latest file affordance.\n- two-turn: Named macro: run two sequential prompts in a new thread.\ndiagnostic:\n- doctor-upload: Named macro: preflight bridge, login, and upload permission remediation.\n- doctor: Preflight browser bridge, login, upload, local files, existing-tab, artifact, localization, report, and selector readiness.\nreport:\n- redacted-run-report: Named macro: create a redacted report for a supplied CommandResult.\n- createReport: Write a durable redacted run report for a command result.\nprimitive:\n- session.bootstrap: Attach to ChatGPT in Chrome and detect login/blocker state.\n- threads.new: Open a new ChatGPT thread.\n- threads.search: Search visible ChatGPT history by query.\n- threads.open: Open a thread by URL, conversation id, title, or search result.\n- messages.compose: Fill the composer without submitting.\n- messages.submit: Submit the current composer contents.\n- messages.ask: Compose, submit, optionally wait, and optionally read.\n- messages.wait: Wait for the latest assistant response to stabilize.\n- messages.readLatest: Read the latest message as Markdown, normalized text, blocks, or HTML.\n- messages.waitAndRead: Wait for completion and read the latest message.\n- artifacts.listLatest: Detect the latest visible generated ChatGPT artifact, such as an image-only result.\n- artifacts.wait: Wait for a visible generated ChatGPT artifact to appear and stabilize.\n- artifacts.downloadLatest: Download or save the latest visible generated ChatGPT artifact.\n- files.preflight: Validate local file paths, size limits, duplicates, zero-byte files, and extension-based MIME/category guesses without opening ChatGPT.\n- files.attach: Attach absolute local file paths through visible ChatGPT upload controls.\n- files.downloadLatest: Download the latest visible ChatGPT file affordance.\n- projects.sources.list: Open or claim a visible ChatGPT Project Sources tab and list source names/statuses without source contents.\n- projects.sources.planAdd: Dry-run an append-only Project Sources file add from explicit local files without opening ChatGPT.\n- projects.sources.add: Append explicit local files to a visible ChatGPT Project Sources list after confirmMutation: true.\n- response.copy: Click Copy response and return clipboard Markdown, with DOM fallback.\n- modes.set: Select a visible model, intelligence, effort, or nested model-version candidate when unambiguous.\n- tools.select: Select a visible ChatGPT tool when unambiguous.",
+  "result": "ChatGPT browser-control SDK commands\n\nworkflow:\n- ask: Ask ChatGPT in a new or selected thread, optionally with files, wait/read, downloads, and reports.\n- askInThread: Open or claim an existing thread by URL, conversation id, title, or search query, then ask and read.\n- askWithFiles: Attach absolute local file paths, optionally set mode, ask, wait, and read.\n- askAndDownload: Ask ChatGPT to produce a visible downloadable output and save the latest exposed file.\n- runMessages: Run sequential prompts where later prompts can use earlier step data.\n- runner.run: Agents-style facade: run a visible ChatGPT browser-control agent against input, files, thread, existing-tab, mode, and response options.\n- responses.create: Narrow Responses-shaped adapter over the visible ChatGPT browser-control runner; rejects unsupported API-only fields before prompt submission.\n- copyLatest: Copy or DOM-read the latest assistant response with Markdown-first fidelity.\n- runPlan: Execute an inline SequencePlan or named macro through the existing sequence engine.\n- new-ask-read: Named macro: open a new thread, ask, wait, and read Markdown.\n- find-open-ask-read: Named macro: search history, open the first match, ask, wait, and read Markdown.\n- find-open-copy-latest: Named macro: search history, open the first match, and copy/read the latest response.\n- attach-ask-read: Named macro: open a new thread, attach files, ask, wait, and read Markdown.\n- ask-and-download: Named macro: ask in a new thread and download the latest file affordance.\n- two-turn: Named macro: run two sequential prompts in a new thread.\ndiagnostic:\n- doctor-upload: Named macro: preflight bridge, login, and upload permission remediation.\n- doctor: Preflight browser bridge, login, upload, local files, existing-tab, artifact, localization, report, and selector readiness.\nreport:\n- redacted-run-report: Named macro: create a redacted report for a supplied CommandResult.\n- createReport: Write a durable redacted run report for a command result.\nprimitive:\n- session.bootstrap: Attach to ChatGPT in Chrome and detect login/blocker state.\n- threads.new: Open a new ChatGPT thread.\n- threads.search: Search visible ChatGPT history by query.\n- threads.open: Open a thread by URL, conversation id, title, or search result.\n- messages.compose: Fill the composer without submitting.\n- messages.submit: Submit the current composer contents.\n- messages.ask: Compose, submit, optionally wait, and optionally read.\n- messages.wait: Wait for the latest assistant response to stabilize.\n- messages.readLatest: Read the latest message as Markdown, normalized text, blocks, or HTML.\n- messages.waitAndRead: Wait for completion and read the latest message.\n- artifacts.listLatest: Detect the latest visible generated ChatGPT artifact, such as an image-only result.\n- artifacts.wait: Wait for a visible generated ChatGPT artifact to appear and stabilize.\n- artifacts.downloadLatest: Download or save the latest visible generated ChatGPT artifact.\n- files.preflight: Validate local file paths, size limits, duplicates, zero-byte files, and extension-based MIME/category guesses without opening ChatGPT.\n- files.attach: Attach absolute local file paths through visible ChatGPT upload controls.\n- files.downloadLatest: Download the latest visible ChatGPT file affordance.\n- projects.sources.list: Open or claim a visible ChatGPT Project Sources tab and list source names/statuses without source contents.\n- projects.sources.planAdd: Dry-run an append-only Project Sources file add from explicit local files without opening ChatGPT.\n- projects.sources.add: Append explicit local files to a visible ChatGPT Project Sources list after confirmMutation: true.\n- response.copy: Click Copy response and return clipboard Markdown, with DOM fallback.\n- modes.set: Select a visible model, intelligence, effort, or nested model-version candidate when unambiguous.\n- modes.get: Read the mode labels shown on the visible composer controls without changing them.\n- tools.select: Select a visible ChatGPT tool when unambiguous.",
   "schemaVersion": "chatgpt.browser_control.backend_response.v1"
 }

package/contracts/v1/fixtures/reports-create-redacted.json

@@ -17,7 +17,7 @@
         "target": {
           "bytes": 335,
           "path": "/tmp/codex-chatgpt-control/reports/contract-fixtures/fixed-contract-report.json",
-          "sha256": "828fb4f5270343f01c60afb9482b22b6b2f524116af54ab91f68d77b7507ffcc"
+          "sha256": "27edafba40f11bcbd8fbda81c36b7caae4f677e4311426086d8bcd3d26aed92a"
         }
       },
       "metaPath": "/tmp/codex-chatgpt-control/reports/contract-fixtures/fixed-contract-report.json.meta.json",

package/contracts/v1/parity-suite.json

@@ -576,7 +576,14 @@
       "surface": "workflows-primitives-diagnostics",
       "fixtures": ["command-descriptors.json"],
       "nodeTests": ["tests/unit/modes.test.ts", "tests/unit/client.test.ts"],
-      "pythonTests": ["tests/test_primitives.py"],
+      "pythonTests": ["tests/test_primitives.py", "tests/test_async_client.py"],
+      "docs": ["references/backend-protocol.md", "references/troubleshooting.md"]
+    },
+    "modes.get": {
+      "surface": "workflows-primitives-diagnostics",
+      "fixtures": ["command-descriptors.json"],
+      "nodeTests": ["tests/unit/modes.test.ts", "tests/unit/client.test.ts"],
+      "pythonTests": ["tests/test_primitives.py", "tests/test_async_client.py"],
       "docs": ["references/backend-protocol.md", "references/troubleshooting.md"]
     },
     "tools.select": {