Issue Station
Sign In
Tools hereToolbox

openclaw - ✅(Solved) Fix [Bug]: [reopen #82344] Regression after upgrading from 2026.5.7 → 2026.5.18: embedded runtime detected as third-party harness while direct Claude CLI still works [1 pull requests, 4 comments, 3 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
openclaw/openclaw#84222•Fetched 2026-05-20 03:42:28
View on GitHub
Comments
4
Participants
3
Timeline
20
Reactions
1
Author
FreeWind6FreeWind6
Participants
clawsweeper[bot]clawsweeper[bot]FreeWind6FreeWind6joshavantjoshavant
Assignees
joshavantjoshavant
Timeline (top)
labeled ×7commented ×4cross-referenced ×3renamed ×2

I've already started the task https://github.com/openclaw/openclaw/issues/82344 and it is marked as completed. But in fact, after upgrading from version 2026.5.7 to version 2026.5.18, the error was repeated.

After upgrading OpenClaw from 2026.5.7 to 2026.5.18, Telegram and Dashboard/UI integrations started failing, while direct CLI usage (openclaw from terminal) continued to work normally with the same Claude CLI auth/session.

Downgrading back to 2026.5.7 restores functionality.

This suggests the issue is not with Anthropic account/auth itself, but with how newer OpenClaw versions handle Claude CLI integration in gateway/UI/Telegram runtime paths.

The failure appears specifically in gateway-mediated interactions, not in direct terminal execution.

Error Message

I've already started the task https://github.com/openclaw/openclaw/issues/82344 and it is marked as completed. But in fact, after upgrading from version 2026.5.7 to version 2026.5.18, the error was repeated. Error in Telegram: ⚠️ API provider returned a billing error — your API key has run out of credits or has an insufficient balance. Check your provider's billing dashboard and top up or switch to a different API key. Error in Logs: LLM request rejected:

Root Cause

I've already started the task https://github.com/openclaw/openclaw/issues/82344 and it is marked as completed. But in fact, after upgrading from version 2026.5.7 to version 2026.5.18, the error was repeated.

After upgrading OpenClaw from 2026.5.7 to 2026.5.18, Telegram and Dashboard/UI integrations started failing, while direct CLI usage (openclaw from terminal) continued to work normally with the same Claude CLI auth/session.

Downgrading back to 2026.5.7 restores functionality.

This suggests the issue is not with Anthropic account/auth itself, but with how newer OpenClaw versions handle Claude CLI integration in gateway/UI/Telegram runtime paths.

The failure appears specifically in gateway-mediated interactions, not in direct terminal execution.

Fix Action

Fix / Workaround

After upgrade to 2026.5.18:

PR fix notes

PR #84374: Fix Anthropic CLI auth routing for shorthand refs

Description (problem / solution / changelog)

Summary:

  • Route Anthropic execution through Claude CLI when the selected Anthropic auth profile resolves to the Claude CLI backend.
  • Preserve shorthand and future Anthropic model refs while seeding canonical Claude CLI runtime policy where known.
  • Add regression coverage for config defaults, migration, runtime aliasing, followup/command execution, and pruning defaults.

Verification:

  • OPENCLAW_VITEST_FS_MODULE_CACHE_PATH=.tmp/vitest-cache-84222-fix-$(date +%s) node scripts/run-vitest.mjs extensions/anthropic/cli-migration.test.ts extensions/anthropic/index.test.ts src/agents/model-runtime-aliases.test.ts src/agents/command/attempt-execution.cli.test.ts src/config/config.pruning-defaults.test.ts src/auto-reply/reply/followup-runner.test.ts src/auto-reply/reply/agent-runner-execution.test.ts
  • AUTOREVIEW_AUTO_TESTS=0 AUTOREVIEW_OPENCLAW_MAINTAINER_VALIDATION=1 .agents/skills/autoreview/scripts/autoreview --mode auto --reviewer auto --fallback-reviewer claude
  • git diff --check

Behavior addressed: Anthropic model refs such as anthropic/opus-4.7 selected with Claude CLI auth could execute via embedded Anthropic instead of the configured Claude CLI runtime. Real environment tested: AWS Crabbox run_a735ed7021d6, lease cbx_5152ee2b4bab, provider aws, type c7a.8xlarge, live Telegram E2E with Convex telegram credential. Exact steps or command run after this patch: Built the synced branch on AWS Crabbox, configured Telegram with anthropic/opus-4.7 and anthropic:claude-cli, installed an absolute fake Claude CLI command, started the gateway, sent a live Telegram message, and asserted routing/output/billing behavior. Evidence after fix: Gateway log included provider=claude-cli; fake Claude CLI was called with --model opus; observed Telegram messages included the CLI marker OPENCLAW_84222_CLI_OK; embedded Anthropic billing trap request count was 0. Observed result after fix: The live Telegram reply came from the Claude CLI route and did not touch embedded Anthropic billing. What was not tested: A real paid Claude CLI account was not used; the live proof used a fake Claude CLI process to make routing and billing isolation deterministic.

Refs #84222

Changed files

  • CHANGELOG.md (modified, +1/-0)
  • extensions/anthropic/claude-model-refs.ts (added, +104/-0)
  • extensions/anthropic/cli-migration.test.ts (modified, +78/-0)
  • extensions/anthropic/cli-migration.ts (modified, +34/-24)
  • extensions/anthropic/config-defaults.ts (modified, +84/-47)
  • extensions/anthropic/index.test.ts (modified, +179/-0)
  • src/agents/command/attempt-execution.cli.test.ts (modified, +55/-0)
  • src/agents/command/attempt-execution.ts (modified, +1/-0)
  • src/agents/model-runtime-aliases.test.ts (added, +84/-0)
  • src/agents/model-runtime-aliases.ts (modified, +79/-1)
  • src/auto-reply/reply/agent-runner-execution.ts (modified, +4/-0)
  • src/auto-reply/reply/followup-runner.ts (modified, +3/-2)
  • src/config/config.pruning-defaults.test.ts (modified, +27/-0)

Code Example

openclaw

---
RAW_BUFFERClick to expand / collapse

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

I've already started the task https://github.com/openclaw/openclaw/issues/82344 and it is marked as completed. But in fact, after upgrading from version 2026.5.7 to version 2026.5.18, the error was repeated.

After upgrading OpenClaw from 2026.5.7 to 2026.5.18, Telegram and Dashboard/UI integrations started failing, while direct CLI usage (openclaw from terminal) continued to work normally with the same Claude CLI auth/session.

Downgrading back to 2026.5.7 restores functionality.

This suggests the issue is not with Anthropic account/auth itself, but with how newer OpenClaw versions handle Claude CLI integration in gateway/UI/Telegram runtime paths.

The failure appears specifically in gateway-mediated interactions, not in direct terminal execution.

Steps to reproduce

  1. Install and configure OpenClaw 2026.5.7

  2. Configure Claude CLI provider according to docs:

    • anthropic:claude-cli
    • OAuth login completed

  3. Verify:

    • Telegram works
    • Dashboard/UI works
    • Direct CLI works

  4. Upgrade OpenClaw to 2026.5.18

  5. Restart gateway/service

  6. Try:

    • sending Telegram message
    • using Dashboard/UI chat

  7. Observe failures/timeouts/errors

  8. Direct CLI usage still works:

    openclaw

  9. Downgrade back to 2026.5.7

  10. Restore previous config/backup

  11. Telegram and Dashboard start working again

Expected behavior

After upgrading from 2026.5.7 to 2026.5.18:

  • Existing Claude CLI auth should continue working
  • Telegram integration should continue responding
  • Dashboard/UI chat should continue responding
  • Gateway should start normally without requiring downgrade flags
  • Behavior should remain consistent with direct CLI execution

Actual behavior

After upgrade to 2026.5.18:

Telegram stops responding Dashboard/UI becomes unusable Gateway/runtime exhibits failures Direct CLI execution still works normally with the same auth/session Error in Telegram: ⚠️ API provider returned a billing error — your API key has run out of credits or has an insufficient balance. Check your provider's billing dashboard and top up or switch to a different API key. Error in Logs: LLM request rejected: Third-party apps now draw from your extra usage, not your plan limits. Add more at claude.ai/settings/usage and keep going.

OpenClaw version

2026.5.18

Operating system

Ubuntu 26.04 LTS

Install method

npm

Model

anthropic/opus-4.7

Provider / routing chain

openclaw cli

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

After upgrading from 2026.5.7 to 2026.5.18:

  • Existing Claude CLI auth should continue working
  • Telegram integration should continue responding
  • Dashboard/UI chat should continue responding
  • Gateway should start normally without requiring downgrade flags
  • Behavior should remain consistent with direct CLI execution
#api#orchestration issue#cache issue#memory leak#API versioning

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

openclaw - ✅(Solved) Fix [Bug]: [reopen #82344] Regression after upgrading from 2026.5.7 → 2026.5.18: embedded runtime detected as third-party harness while direct Claude CLI still works [1 pull requests, 4 comments, 3 participants]