Persistent HTTP 400 "Invalid project resource name projects/" — Empty cloudaicompanionProject after ESET firewall removal (Windows 11, Antigravity 1.22.2, AI Pro)

Google Antigravity
, , , ,
1

Summary

Every Agent run fails immediately with HTTP 400 INVALID_ARGUMENT because the cloudaicompanionProject is empty (projects/). Extensive local debugging has confirmed this is a server-side provisioning failure: the loadCodeAssist API never returns a cloudaicompanionProject, and onboardUser returns done: true without a project binding. Requesting a server-side reset of the cloudaicompanionProject binding for this account.

Root Cause Analysis (Detailed)

Through systematic debugging I have identified the full causal chain:

  1. Original blocker (now resolved): ESET Endpoint Security 12.0 kernel-level WFP drivers (epfw.sys, epfwwfp.sys) were intercepting and aborting localhost TLS handshakes between Antigravity’s extension host and its Go language server. This caused wsarecv: An established connection was aborted by the software in your host machine errors during every startup, which prevented the OAuth token flow from completing.

  2. ESET has been fully uninstalled. Post-removal logs confirm zero TLS handshake errors — the localhost communication is now clean.

  3. However, the account provisioning state is now stuck server-side. The loadCodeAssist endpoint at daily-cloudcode-pa.googleapis.com/v1internal:loadCodeAssist returns HTTP 400 with "Request contains an invalid argument" because the request contains an empty project resource name (projects/). Inspection of Antigravity’s internal state database (state.vscdb) confirms that the enterpriseGcpProjectId field in antigravityUnifiedStateSync.enterprisePreferences is empty (protobuf value GgA= = zero-length string).

  4. The onboarding flow does not re-provision the project. Resetting onboarding, signing out/in, and switching between email aliases all fail to trigger project creation. The server appears to consider onboarding already complete.

  5. User-created GCP projects are not accepted. During onboarding, manually selecting a GCP project with Cloud AI Companion API enabled) results in “This Google Cloud project is not supported.”

Environment

  • OS: Microsoft Windows 11 Version 24H2 (OS Build 26100.7840), x64
  • Antigravity: 1.22.2 (updated from 1.16.5 during troubleshooting)
  • Electron: 39.2.3 → updated with new version
  • Device management: Not work-managed (personal device, no MDM)
  • VPN/Proxy: None active (NordVPN installed but fully disabled/quit; Cisco AnyConnect present but not connected)
  • Account type: Personal Gmail, Google AI Pro subscriber
  • 2FA: Enabled

Correlation IDs

Most recent failure (April 8, 2026):

  • Trajectory ID: 1d9820d8-d54f-
  • TraceID: 0xdc57bc
  • X-Cloudaicompanion-Trace-Id: dc57bcde
  • Timestamp: 2026-04-08 15:27:38 UTC

Earlier failure (February 16, 2026, pre-ESET removal):

  • Trajectory ID: 71e77778-d804-4
  • TraceID: 0x212edb1
  • X-Cloudaicompanion-Trace-Id: 212edb16

Error Payload

{
  "error": {
    "code": 400,
    "details": [{
      "@type": "type.googleapis.com/google.rpc.BadRequest",
      "fieldViolations": [{
        "description": "Invalid project resource name",
        "field": "name"
      }]
    }],
    "message": "Invalid project resource name projects/",
    "status": "INVALID_ARGUMENT"
  }
}

Relevant Log Excerpts (April 8, 2026, post-ESET removal)

Server startup — no TLS errors (ESET fix confirmed):

I0408 13:36:11.660482 server.go:1235] Starting language server process with pid 33588
I0408 13:36:11.673739 server.go:440] Language server will attempt to listen on host 127.0.0.1
I0408 13:36:11.678432 server.go:454] Language server listening on random port at 62223 for HTTPS (gRPC)
(Antigravity) 2026-04-08 13:36:12.889 [INFO]: Language server started
I0408 13:36:12.933051 server.go:1679] initialized server successfully in 1.2695893s

loadCodeAssist fails immediately with 400:

I0408 13:36:13.633669 http_helpers.go:123] URL: https://daily-cloudcode-pa.googleapis.com/v1internal:loadCodeAssist Trace: 0x6664e81b40e89072
I0408 13:36:13.732904 http_helpers.go:166] Failed to make code assist backend request (fetchAdminControls): {
  "error": { "code": 400, "message": "Request contains an invalid argument.", "status": "INVALID_ARGUMENT" }
}

After onboarding reset — OAuth token not found:

E0408 13:38:07.541393 server.go:545] Failed to get OAuth token: error getting token source from auth provider: state syncing error: key not found
W0408 13:38:07.541393 client.go:167] failed to set auth token
E0408 13:38:15.151162 interceptor.go:74] /exa.language_server_pb.LanguageServerService/GetUserStatus (unknown): error getting token source: state syncing error: key not found

Agent planner request — empty project:

I0408 10:27:16.174261 planner_generator.go:283] Requesting planner with 7 chat messages at model retry attempt 1 and API retry attempt 1
E0408 10:27:16.551483 log.go:398] agent executor error: INVALID_ARGUMENT (code 400): Invalid project resource name projects/

Internal State Verification

Queried state.vscdb → antigravityUnifiedStateSync.enterprisePreferences:

  • Field enterpriseGcpProjectId is present but contains an empty protobuf string (base64: GgA=, decodes to \x1a\x00)
  • antigravityAuthStatus contains a valid access token (ya29.*) and correct account name
  • antigravityUnifiedStateSync.userStatus contains no project references

Troubleshooting Already Attempted (All Failed to Resolve)

Network/security fixes (resolved TLS errors but not provisioning):

  • :white_check_mark: Full uninstall of ESET Endpoint Security 12.0 (including kernel drivers epfw.sys, epfwwfp.sys) — confirmed removed via driverquery
  • :white_check_mark: Disabled and quit NordVPN + Threat Protection (all kernel drivers: NDivert, NordDivert10, tapnordvpn)
  • :white_check_mark: Set HOME environment variable to %USERPROFILE%
  • :white_check_mark: Set GOOGLE_CLOUD_PROJECT environment variable

Auth/state resets (did not trigger re-provisioning):

  • :white_check_mark: Deleted .antigravity and %APPDATA%\Antigravity folders, re-onboarded
  • :white_check_mark: Ctrl+Shift+P → “Antigravity: Reset onboarding” (multiple times)
  • :white_check_mark: Revoked Google account connections and re-authenticated
  • :white_check_mark: Signed in with both primary Gmail
  • :white_check_mark: Cleared GPU cache, .gemini folder, and Antigravity cache
  • :white_check_mark: Cleared Windows Credential Manager of all Antigravity/Google/Gemini entries
  • :white_check_mark: Fixed UTF-8 BOM corruption in mcp_config.json files (caused MCP parse errors)
  • :white_check_mark: Disabled all MCP servers

GCP project configuration (not accepted):

  • :white_check_mark: Created new GCP project and enabled Cloud AI Companion API
  • :white_check_mark: Enabled Cloud AI Companion API on existing project
  • :white_check_mark: Enabled 2FA on Google account
  • :white_check_mark: Attempted to select GCP project during onboarding — rejected as “not supported”
  • :white_check_mark: Set cloudcode.project in settings.json — ignored by language server

Cannot test with alternate account: Only other Google account is ASU Workspace/managed and ineligible for Antigravity.

Request

Please reset the server-side cloudaicompanionProject binding for account so that the next login triggers a fresh project auto-provisioning. The original provisioning failure was caused by ESET firewall interference (now resolved), but the account state is stuck server-side with onboardUser: done and no project assigned.

This matches the known pattern described in:

  • gemini-cli#24747 (filed April 6, 2026): “stale cloudaicompanionProject binding on the server”
  • CLIProxyAPI#1030 (filed January 15, 2026): “loadCodeAssist API doesn’t return cloudaicompanionProject”
  • Multiple forum threads dating to February 2026

Related Issues

2

Same issue here — Google AI Pro subscriber

Exactly the same symptoms:

  • HTTP 400 “Invalid project resource name projects/”
  • cloudaicompanionProject always empty
  • loadCodeAssist returns INVALID_ARGUMENT
  • onboardUser returns done: true with no project binding

My latest Trace ID: 1148000f-efcc-4d85-a836-1fecc57
(also have Trajectory ID from Cockpit Tools)

Already tried everything in this thread + clean reinstall, new GCP projects, 2FA, delete %APPDATA%\Antigravity, etc.

Requesting server-side provisioning reset for my account as well.

Thanks for the extremely detailed write-up!