Claude Code. Any Model. The most powerful AI coding agent now speaks every language.
docs(reports): add the code-mode bench run and V3 credit block
Jack Rudenkob70e213d163b
diff --git a/ai-docs/reports/codex-tool-calls-and-code-mode-2026-10.md b/ai-docs/reports/codex-tool-calls-and-code-mode-2026-10.mdindex 7a14cbca..e4f9f56f 100644--- a/ai-docs/reports/codex-tool-calls-and-code-mode-2026-10.md+++ b/ai-docs/reports/codex-tool-calls-and-code-mode-2026-10.md@@ -1,6 +1,6 @@# Codex tool calls: zero-delta fix, parallel-wrapper retry, code mode (2026-10)-Branch `worktree-investgiage-gpt-file`, base `086cf608`, 69 commits. Sibling report:+Branch `worktree-investgiage-gpt-file`, base `086cf608`. Sibling report:`codex-parallel-wrapper-degeneration-2026-10.md` (the request-factor experiment, part B).## The defect@@ -73,13 +73,16 @@ request shape from the design, not from the harness.## Open items-- The commits are unsigned (1Password signing was unavailable); re-sign before any push and reword:- `eefa4f8a` (89 chars), `039d3b44` (77), `a2452d0d` (73), `932f27e1` (74); scope `code-mode` →- `openai-codex` on `c804ae84`, `88561b8d`.+- Re-run check: `ai-docs/benches/codex-code-mode/` (madbench; `classic`, `auto`, `auto-edit` on+ `[email protected]`, subscription only). First run 2026-10-20 on Claude Code 2.1.304: 15/15, 19/19 exec+ scripts resumed, 0 input errors in tool_results. Needs `--concurrency 1` (cause unknown). Claude Code+ 2.1.303, where D1 lived, was not pinned.- Behaviour changes outside code mode, for the PR and release notes: `d6629064` identity-filters`instructions` on every `cx@`/`openai@` Responses request; a 401 whose forced refresh lands on anotherChatGPT account is answered with the upstream 401, not resent (`1fe1b580`, `a808fab2`).-- `openai@` regression check not run (no API credits).+- `openai@` regression check (V3) not run: on 2026-10-20, after a credit top-up, the Keychain+ `OPENAI_API_KEY` (`sk-proj-…`) still got `credit_balance_exhausted` inside a 200 stream; claudish+ reached `https://api.openai.com/v1/responses` on the `direct-api` arm as intended.- Not proven live: a mixed classic/code-mode history after `--resume`; Esc on a nested permission prompt.- Not covered by tests (no seam or no capture): the real OAuth refresh single-flight; `codeMode` through aprovider profile; an image tool_result inside an exec output.