TRIAD Codex Dispatch 0.2.561

TRIAD Codex Dispatch 0.2.561

Explicit v2 JSON, verdict bindings, roster resolution and prompt rendering now use the bundled triad-dispatch-lib 0.1.0.dev3 from reviewed source 2d1e62e7de6f33fa13a918a39579c6b5e74def00. Independent pins verify the complete package, license and resource bundles. Runtime use needs no separate library checkout or download. Native provider adapters, authentication, classifiers, spawning and recovery remain host-owned; the shared review scheduler is not adopted.

The bundled update settles callbacks created before overflowing wait setup and retains the concurrent CPython cache-publication correction. Host prompt-path encoding preserves CR/LF, quotes and non-ASCII filename identity. Compatibility-resource regressions cover the retained host assets. Missing or unexpected bundled paths are reported with their verifier cause and bootstrap recovery guidance, preserving conflicting files. See the changelog and refresh record.

Use the supported update procedure, satisfy the shipped requirements in your selected Python 3.12+ environment, run the newly installed bootstrap from your target workspace, and start a fresh Codex session. Prepare a new review ID and basis after updating. Legacy formal metadata and web-clause helpers also require the verified bundled parser; restore exact package content before retrying an integrity refusal. Claude's independent adoption handoff and the Gemini Argon Google-route handoff retain separate host and route boundaries.

Reviewed commit: c81c02984825af7dd1ec69a14af83f0ff18d0551; tree: 9cf7e01b82ef071b4922fede47cbb813a34778b9. Native CI run 38665062677 passed on both platforms. Its checkout 0c54812517c5829cd09ebd2fb0e0881f88ffc39f is the synthetic merge of this candidate into bd96cdbf0eb83e2187b2b7fa3ac9cd701b58db7e, with the identical tracked tree.

Native runner	Runtime	Source tests	Extracted archive tests
Ubuntu 24.04 x86_64	Python 3.12.15	2,555 passed, 6 skipped in 676.65s	2,555 passed, 6 skipped in 707.51s
macOS 15.7.9 arm64	Python 3.12.10	2,557 passed, 4 skipped in 944.35s	2,557 passed, 4 skipped in 891.98s

Four existing skips cover AGY JSON-schema transport and Gemini model/effort contracts; Ubuntu also skips two case-insensitive filesystem cases. Both source and archive isolated runtime checks validated exact bundled module origins, all four routes, 24 binding refusals, three overflow refusals, one duplicate refusal and no runtime skips. Their temporary environments used a parent path containing Hangul, quotes and a literal backslash-n; the reported environment removal and child absence were checked before each result was printed. Both jobs verified 80 packaged hash targets.

The uploaded triad-codex-dispatch-0.2.561.tar contains all 373 tracked files with bytes and modes matching the reviewed commit, including the 23 pinned package/license files. Its SHA-256 is b066b1756b0486b7d9b494e517e86a9e5a43eca31efd69085fe6b6470e1dc996. The public release assets were downloaded again: the tar is byte-identical to the prepared artifact, and SHA256SUMS and GitHub's asset digests agree. The public tag targets the reviewed commit above. The CI synthetic-merge archive has a separate hash, 69e655d0f4e55bc87abc1751de596f3a3b3a68d1b24d2940d719e3fcb30ba21d; it is not the uploaded release archive.

Final round 20261015-codex-library-r5-69ca5c returned SAFE from all four selected legs. Canonical collection returned AGREED with final integrity checks passing; the closed round was exported. PR #44 retains two nonblocking limits: the internal schema-loader import works under the current exact pin but needs attention during future refreshes; a path containing the exact canonical caveat sentence can suppress the separate residual caveat paragraph for Claude/Codex. The residual remains encoded and fenced; provider or security impact has not been observed. No general filename exclusion is claimed.

These results establish reviewed source, native CI, published archive identity and downloaded-asset equality. Local installation and fresh installed-session exposure are separate workstation checks. These results do not establish another host's adoption, authenticated-provider success or effective company policy.