[EvilWeasel]

The Windows startup check opened the installer as its embedded application 7z archive, then looked for the NSIS runtime path $PLUGINSDIR/app-64.7z inside that archive. The filter selected zero files and the payload presence check correctly stopped the job before app startup.

Extract that embedded 7z directly from the exact selected installer into a fresh, unique scratch directory. Require Morrowbyte.exe to be a file before hashing and running the existing desktop smoke with the expected version. The Builder binary resolver, actual launch/version checks, process cleanup and diagnostics remain in place.

Validation:

• Reproduced the original zero-files/payload-missing failure with the unchanged workflow and an app archive produced by the pinned Builder NSIS archive routine and differential options, wrapped in the observed prefix/tail lengths.
• The actual corrected PowerShell step extracts the known executable and app-resource bytes. A stale scratch app is ignored; an executable directory, corrupt archive and failed launch boundary still fail and clean up. This is a PowerShell 7/Linux format probe with synthetic prefix/tail bytes, not original Windows installer or app-startup proof.
• All 28 existing Nightly contract tests, the normal exact-base integration check and staged format/lint hooks passed. Parsed workflow comparison confirms that the edit is confined to the Windows startup run.

Model: GPT-6.1 Sol. Harness: Codex.

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-13T16:32:12.704322Z	c19706b	PR opened

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.