Only this prompt:
We are getting good test now but have bit of hiccup like stopping midway that's strange here is the result of the test: Deep test complete — all green Ran a full provider validation pass against the live Cursor backend and fixed two bugs found along the way. Test results LayerResult Unit tests125 passed, 2 skipped (live probes) pnpm checkPass (format, lint, typecheck, test, build) Live catalog probe59 models discovered, 4 in picker, effort controls on Grok/Codex Live tool-call probeComposer emits tool-call block for run_code in ~3s Live smokeText round-trip OK, shell→bash tool path OK Deep integration (scripts/deep-test.ts)All 6 checks PASS Bugs fixed src/core/h2-stream.ts — HTTP 401/403 responses hung until idle timeout instead of failing immediately. Now awaits response headers after stream ready and throws rpc failed with HTTP {status} before frame parsing. src/cursor-adapter.ts — Tool suspend race: when the harness breaks on finish{tool-calls}, generator.return() ran discardActiveRun() before suspended was set, killing the open HTTP/2 stream. Phase-2 resume then failed with CURSOR_STREAM_FAILED. Fix: set suspended = true before yielding the finish chunk. New coverage tests/cursor-adapter.spec.ts — resumes the same active run after the consumer breaks on tool-calls finish tests/README.md — ledger entries for both fixes scripts/deep-test.ts — opt-in live integration script (not CI; run with system Cursor auth) How to re-run live tests sh Copy # Full unit suite pnpm test # Live probes (needs ~/.config/cursor/auth.json or CURSOR_ACCESS_TOKEN) DSH_CURSOR_LIVE_PROBE=1 pnpm vitest run tests/live-probe.spec.ts tests/live-tool-probe.spec.ts # Comprehensive provider smoke pnpm exec tsx scripts/deep-test.ts pnpm exec tsx scripts/live-smoke.ts Provider capabilities verified Model listing + resolveModel (effort variants for Grok/Codex) Plain text streaming (Composer 2.5) MCP tool-call surfacing (run_code, bash) Full tool loop resume on the same bidirectional Run (call → suspend → host result → continue) Connect streaming media type + framing Transient truncation retry ladder Note: tests/tmp-diag.spec.ts and tests/tmp-h2-probe.spec.ts are leftover diagnostic probes (30s timeout each) — consider removing or gating behind an env flag if you want faster CI. Context injection dsh-factory Factory completion Think Work is complete. I will call `factory_finish` with the verified outcome and a concise summary. Code Discover factory_finish tool schema Context injection dsh-fovea Fovea detected repository drift Think Calling factory_finish with the verified outcome. Code Settle Factory task as succeeded Tool call factory_finish · succeeded Think The factory_finish call was accepted. Factory task settled: succeeded — deep provider testing complete, two bugs fixed, full gate and live probes verified. ===== look on dsh-codex I guess for more comparison since, it's quite similar
It's official! @ChatGPT@OpenAI
Codex nerf... Sol doesn't consume this much even on plus.
Also I have a 95% cache so don't bring that shit again!
Only got 1 prompt and poof! I am used to with the usual openai nerfing practices and such but this SUCKS hard!
Isn't sloppy content just an irk...
Content and design are kinda subjective.
Looks bad = irk
Written badly or full of bullshit = irk x2
AI slop on top of that = irk x3
So yeah, edit your AI slop. Be irk-less.
Attention @jetbrains users!!
https://t.co/0r527T1oUW now runs inside your IDE. pi-acp-jetbrain is an ACP adapter for IntelliJ 2026.2+, putting pi in the agent panel with IDE tools like inspections wired in.
Install: npm i -g pi-acp-jetbrain
@openai your customer service sucks!! I got this suspicious thing, I did remove all device and did advance security and still the same. 200$ plan and I can use it.