Skip to content

CLI v0.6.0: reliability, material contracts and native Windows validation - #8

Merged
cyaolong merged 10 commits into
mainfrom
codex/cli-v060-reliability
Oct 23, 2026
Merged

cyaolong merged 10 commits into
mainfrom
codex/cli-v060-reliability

Conversation

@cyaolong

@cyaolong cyaolong commented Oct 22, 2026 •

Copy link
Copy Markdown
Contributor

This candidate separates the CLI runtime, preserves structured backend material errors and metadata, adds all four Berry models (16 models total), and hardens generation/upload recovery, update notifications, downloads, credentials, and Windows installation. Material compliance remains server-owned; the backend is unchanged.

Adversarial fixes cover confirmation-time price changes, active-operation purge protection, truncated/partial downloads, BOM request files, damaged/foreign DPAPI credentials, Windows cancellation and delayed cleanup, and preservation of a launcher replaced during uninstall. The test harness also rejects missing/stale evidence, false success exits, leftover processes, and platform-dependent source bytes.

Runtime candidate: af62761bf4e3f50bda5f0911620554caca0eb856. Documentation-only acceptance update: dd564924a54a9d6f809b7933e5d6fcbd0527add9; shipped runtime, tests, installers and candidate contract fingerprint are unchanged. CI run 39945245085 passed all 12 validation jobs. Release jobs were intentionally skipped.

  • Windows Server 2025 x64 / Python 3.10 and 3.14; Server 2022 x64 / 3.10; Windows 11 ARM / native ARM64 3.14: each passed 487 base tests, 54 mandatory native cases, and 13 installed HTTP workflows. The 34 base skips are explicitly mapped: 31 require native equivalents, two are Unix-only, and one is the separate website check.
  • Native evidence proves an ordinary user token, PowerShell 5.1/7/CMD behavior, real ConPTY input/cancellation, immutable v0.5.0 upgrade, DPAPI, file locks, rollback and actual delayed cleanup. The supervisor's real exit probes return 0/37/130; all four final cleanups have no surviving or forcibly killed processes.
  • macOS and Linux / Python 3.10 and 3.14: each passed 518 base tests, with three explicit platform/external skips. Linux 3.14 also passed the 13 HTTP workflows and all 10 behavioral mutation checks.
  • Local Python 3.14 passed the full adversarial profile, five-language website examples, immutable Generate contract at 569aa948d5ea70c15ce719783e71935d02309743, real baseline upgrade, QR decoding and official Skill validation. Implementation and test validity received independent review.

For the af62761 offline run, all eight matrix artifacts and the local report have identical source fingerprint e757ebf0652cef6c158a8bd420041a46fb140d826432f759a2113f91a60ef388. Downloaded artifacts were checked against GitHub SHA-256 digests. Candidate contract fingerprint: 166c866bd575406364cfc45339bb9041025b56a7131b45ebbf92306b0f16db12. Windows now checks the same recorded contract evidence directly.

See automation instructions and findings and fixes.

Live acceptance now reuses the existing authorized person under the owner's explicit scope. Exactly one existing HEIC image (989,564 bytes) was uploaded with one presign, one object PUT and one group registration. CLI and MCP both reported UPLOADED_TO_ARK / ready=true; group listing grew by one and retained the original asset; local upload state is completed; downloaded bytes match the original SHA-256; the account balance is unchanged. No authorization session, paid generation, rename or deletion was performed. The new test asset is retained in the existing group. Full release acceptance and 29 acceptance/contract tests pass. See redacted live evidence. No private asset IDs, URLs, filenames or account credentials are included in public evidence.

Remaining boundaries: new QR/liveness verification and paid generation were not performed. Windows 10/11 x64 desktop validation is explicitly excluded by the owner. Network-filesystem combinations are not exhaustively covered. The HEIC metadata state was PROCESSING, then PENDING at the last recorded observation (the original asset was FAILED); readiness does not assert compliance with every model, and no metadata retry was requested. Installation prefixes containing % or ! are explicitly rejected; special request filenames use the documented shell form. Price is rechecked before submission but is not an atomic backend quote reservation. The recorded private-backend contract is not a live remote check.

The documentation update also passed all 12 validation jobs in CI run 39966176216; all three publication jobs were skipped. Full local release acceptance passed against the final record without changing the gate. Only acceptance documentation changed in dd56492.

The owner explicitly approved formal v0.6.0 publication on 2026-10-23, including the existing release-note update and automated installation/update verification. Commit 9d6a24f finalizes the existing versioned release notes and records approval; it does not change runtime, installers, tests or the candidate contract fingerprint. Publish only after the normal immutable Draft and native installation gates pass.

@cyaolong cyaolong changed the title refactor: harden CLI workflows and automate update notifications feat: harden CLI workflows and sync the 16-model material contract Oct 23, 2026
@cyaolong cyaolong changed the title feat: harden CLI workflows and sync the 16-model material contract CLI v0.6.0: reliability, material contracts and native Windows validation Oct 23, 2026
@cyaolong
cyaolong marked this pull request as ready for review October 23, 2026 14:16
@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 23, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-23T14:20:44.637830Z 9d6a24f Draft marked ready
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@cyaolong
cyaolong merged commit 6b53ef7 into main Oct 23, 2026
15 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant