Historical version
This page records design, operation, or validation facts from an earlier release. It is not current V7.17.3 guidance. Start from the documentation hub for current instructions.
V7.13.4 Release Notes¶
V7.13.4 fixes Windows installation permissions and delegation entry points in Codex Desktop. The product supports the Desktop app only. Existing stable-component records remain internal contract-regression references; they do not define standalone CLI support or Desktop version admission.
- Preserve the destination parent's read access when installing public Plugin, Hook, and runtime payloads without adding write permission. Private state and restoration copies retain their existing rules.
- Exclude generated Python bytecode so public runtime deployment matches the release payload.
- Normalize bare, dotted, and Desktop-concatenated delegation tool names, and register complete names in one Pre/Post handler group to avoid missed or duplicate handling.
- Preserve Windows path identity, recovery safeguards, frozen V3 defaults, and GPT-5.6 compatibility. The nine GPT-6 candidates remain subject to scenario qualification and real evidence.
Actual Desktop installation, ordinary-sandbox reads with writes denied, and V4 message-reentry rejection were verified. Positive creation still fails before execution with V4_REQUEST_MESSAGE_MISMATCH: this host supplies an opaque encrypted message without a verified plaintext-binding interface. A controlled plaintext-rewrite experiment also produced a decoding error and was not adopted. This release does not bypass validation, synthesize receipts, enable GPT-6 defaults, or claim model qualification or positive native admission.
Existing component samples from 0.157.0 through 0.153.0 and their official artifact/source digests remain internal references. They are not the Desktop app's supported-version window; acceptance must use the actual running Desktop component.
Commit, push, installation, CI, tag, Release, assets, provenance, and public effectiveness require separate readback. Full model-performance evaluation, reasoning-effort gains, and default migration remain unfinished follow-up work.