Skip to content

TRIAD Codex Dispatch 0.2.563

Choose a tag to compare

View all tags
@chaniri01 chaniri01 released this 23 Oct 07:30
v0.2.563
89acf3a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

TRIAD Codex Dispatch 0.2.563 accepts AGY provider metadata without weakening verdict validation, switches the default AGY formal route to Argon/high, and supports an explicit 2400-second legacy review deadline.

  • Allows --timeout 2400 in the three legacy provider wrappers. Existing Claude 1200-second and Google 600-second calls remain supported; other legacy timeout values remain refused. The AGY child receives 2390 seconds. Raw and public-v2 defaults are unchanged.
  • Separates undeclared top-level toolAction and toolSummary metadata before AGY result validation while preserving original output. Declared application fields and aliases retain their validation. Strict original JSON, duplicate/non-finite rejection, verdict bindings, terminal failures and bounded recovery remain enforced.
  • Rejects malformed terminal output wrapped in non-JSON Unicode whitespace, preventing an earlier SAFE payload from replacing a later terminal result. The regression is reproduced synthetically; occurrence in actual vendor output has not been observed.
  • Uses gemini-4-argon-high with high for default AGY and fixed legacy review. Argon/Flash guidance is synchronized. Explicit raw/public-v2 model overrides and the separate Gemini CLI route remain supported.
  • Bundles published triad-dispatch-lib 0.1.0.dev5 at 859f27c07c1123e87b50344f587999a0cbc172a6, with release-wheel SHA-256 833d55e9bf394885423bf1ed954a972747e1968b52b911b5253a9b7971512c7f and all 26 package/license identities pinned. No separate library installation is needed.

Formal review round 20261023-codex-agy-timeout-r5-91a7d3 closed with four SAFE verdicts (Claude, Argon, Flash and fresh native Codex), ROUND_INTEGRITY_OK, and ADMITTED_SAFE. Non-blocking observations and disposition are recorded in PR #47. The separate damaged-inventory artifact-persistence limitation remains.

Exact reviewed candidate: facbaaf98c9b74ee7b8f8753f3e48175ab7b273d.
Merged main/tag target: 89acf3a7c3deda49d0e48942020e5061e638aa61 (identical tracked source tree).
Archive: triad-codex-dispatch-0.2.563.tar.
SHA-256: aa18c6de19bbaa672548e028af1a30d639f39ce1a4e1525f728c6ab05e0c4f9a.
Every archive member was compared with its committed source bytes and executable mode (379 files).

Local macOS/Python 3.12.13 full suite: 2628 passed, 5 skipped. Native CI run 39912745899 passed source, clean committed archive and isolated bundled-runtime checks on macOS and Ubuntu against the same source tree. Source and archive each passed 2628 tests with 5 skips on macOS arm64/Python 3.12.10, and 2626 tests with 7 skips on Ubuntu x86_64/Python 3.12.15. CodeQL also passed.

Follow the Codex update procedure, rerun bootstrap from the returned installed cache, and start a fresh Codex session. If another host owns the four shared launcher names, choose their ownership or a separate supported launcher directory before applying bootstrap; do not overwrite another host's launchers blindly.

The tag-pinned changelog retains candidate and earlier-round status prose frozen before the final R5 review. This release record supplies the final publication and verification status. Source/archive checks, local installation, fresh-process skill exposure, and authenticated provider behavior remain distinct evidence.

Assets 4
Loading