PR https://git.cafe/pingdotgg/t3code/pulls/35963 PR head c5121e456a806637fd99128858e6f3e91099752d Baseline github/main d0a754daeaa972101881befe74a6f99e24e25da7 Linux; repository Playwright 1.64.1; /usr/bin/google-chrome Main: vp i; vp run dev in /home/user/main (background, captured PID 16967) [dev-runner] serverPort=16733 webPort=8693 baseDir=/home/user/main/.t3 App-created project /home/user/usage-verification, real git repository. App-added registry instance acpRegistry_glm, glm-acp-agent 1.17.1. Usage > Limits: No provider on the selected environments reports subscription limits. Codex gpt-6.1-sol read README.md and git branch: heading usage verification, branch master. Stopped main process group captured at launch before starting PR. PR: vp i; vp run dev --home-dir /home/user/t3code/.t3 --no-seed (background) [dev-runner] serverPort=13773 webPort=5733 baseDir=/home/user/t3code/.t3 Same real repo and GLM agent, configured through the app. Usage > Limits and thread /usage-limits: Set Z_AI_API_KEY or run glm-acp-agent --setup to check quota. Persisted reading: windows=[], unavailable.reason=unsupported, group.id=acpRegistry:acpRegistry_glm, group.label=GLM Agent. Agent sign-in confirmation persisted authenticated=false. Codex gpt-6.1-sol completed real README and branch turn in 7.8s. Codex /usage-limits: This account has no subscription limits. (proxy account) Failed GLM turn, no manual refresh during this check: Before checkedAt: 2026-11-18T10:08:09.978Z Client prompt: Reply with ok. Do not change any files. Client error: No API key found. Set the Z_AI_API_KEY environment variable, or run glm-acp-agent --setup to store one. Server: [10:08:50.174] WARN orchestration-v2.acp-prompt-failed After checkedAt: 2026-11-18T10:08:50.399Z Fresh unsupported reading replaced prior unsupported reading after real prompt failure. Blacksmith: 224 existing tests passed across six files: zaiUsageLimits.test.ts, driver.test.ts, managedProvider.test.ts, provider-core/usageLimits.test.ts, shared/usageLimits.test.ts, effect-acp/client.test.ts. Not verified: authenticated Z.ai quotas, monthly labels/pace markers, exhaustion/fallback, BigModel endpoint, native mobile, Electron shell, remote/relay/tunnel. No Z.ai credentials are installed. No mocks, fixture quotas, or demo turns used. Manual Usage > Refresh limits: Before checkedAt: 2026-11-18T10:09:36.941Z After checkedAt: 2026-11-18T10:09:47.650Z Missing-credentials state stays visible; new agent reading is published. Neighboring Usage Cost and Tokens tabs both render and Limits remains reachable.