fix(codex): make package timestamps timezone-independent

[SeuPut0705]

Who is submitting this PR? (required)

Field	Value
Your model + version	OpenAI GPT-5.6-sol (primary and final review); GPT-5.6-terra (implementation and task review)
Harness + version	ChatGPT/Codex desktop 26.715.31925 (build 5551), embedded Codex CLI 0.145.0-alpha.18; local Codex CLI 0.144.4
All plugins installed	documents, pdf, spreadsheets, presentations, template-creator 26.715.12143; sites 0.1.30; browser and chrome 26.715.31925; computer-use 1.0.1000451; visualize 1.0.12; slack, game-studio, github, codex-security 2f1a8948; active Superpowers skill bundle 6.1.1
Human partner who reviewed this diff	@SeuPut0705

What problem are you trying to solve?

Running tests/codex/test-package-codex-plugin.sh on macOS in Asia/Seoul produced two timestamp failures. ZIP entries were stored as (1980, 1, 1, 9, 0, 0) instead of midnight, and the tar assertion compared a localized date string (Dec 31 1969) with macOS bsdtar's Jan 1 1970 display.

Direct reproduction showed that UTC and Asia/Seoul ZIP packages had different SHA-256 values and stored 00:00 versus 09:00, while both tar packages were byte-identical and had raw member mtime values of 0. The ZIP command inherited the host timezone even though the preceding touch ran under UTC; the tar test was inspecting presentation text rather than archive metadata.

What does this PR change?

Run Info-ZIP with TZ=UTC, add a regression test that compares real ZIP bytes produced under UTC and Asia/Seoul, and inspect tar member mtimes through Python's tarfile API instead of localized tar -tvf output.

Is this change appropriate for the core library?

Yes. This fixes the repository's Codex packaging script and its integration test without adding dependencies or changing skill behavior. Deterministic portal packages benefit every maintainer producing the official Codex archive.

What alternatives did you consider?

• Keeping UTC only on touch: this is the current behavior and leaves Info-ZIP's DOS timestamp conversion timezone-dependent.
• Relaxing the ZIP timestamp assertion: this would hide non-reproducible package bytes.
• Rebuilding ZIP output with Python: this would replace the existing production archive path for a one-environment-variable fix.
• Comparing tar -tvf under a forced locale/timezone: the rendered date remains tool-specific; raw member.mtime is the portable archive contract.

Does this PR contain multiple unrelated changes?

No. The one-line packaging fix and both test changes address the same archive timestamp determinism defect.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art.
• Related PRs: none found for package-codex-plugin timestamp, archive timezone deterministic, or Codex package zip in open/closed PRs and issues.

Environment tested

Harness	Harness version	Model	Model version/ID
ChatGPT/Codex desktop on macOS 26.5.2, Asia/Seoul	26.715.31925 (build 5551), embedded Codex CLI 0.145.0-alpha.18	OpenAI	GPT-5.6-sol / GPT-5.6-terra
Shell integration test: bsdtar 3.5.3, Info-ZIP 3.0, Python 3.14.6	local Codex CLI 0.144.4	N/A	N/A

New harness support

Not applicable. This PR does not add or change harness support.

Evaluation

• Initial prompt: the human partner approved proceeding with a PR after the KST packaging test failure was reproduced during Codex package verification.
• Before: explicit UTC and Asia/Seoul ZIP builds had different SHA-256 values and timestamps; the focused test reported two failures.
• After: the focused packaging integration test passes, including byte-identical UTC/Asia/Seoul ZIP output and raw tar mtime == 0; the marketplace manifest test and git diff --check also pass.
• Model eval sessions after the change: 0. This is packaging infrastructure, not behavior-shaping skill content.

Rigor

• This is not a skills change; superpowers:writing-skills is not applicable.
• The change was tested adversarially with explicit UTC and Asia/Seoul environments.
• No carefully tuned behavior-shaping content was modified.

Human review

• A human has reviewed the COMPLETE proposed diff before submission.