Best possible solution:

Keep the remaining work on #95189, land one cached helper covering all three runtime prompt producers, then close overlapping PRs and the linked issue after it reaches main.

Do we have a high-confidence way to reproduce the issue?

Yes. Source inspection shows current main and v2026.6.9 still use os.type() plus os.release() in all three runtime prompt producers, and the linked issue supplies macOS product-version versus Darwin kernel-version evidence.

Is this the best way to solve the issue?

No. This branch is plausible, but the best remaining path is the proof-positive sibling at #95189 because it covers the same runtime prompt producers with live macOS proof and stronger cache/test coverage.

Security review:

Security review cleared: No concrete security or supply-chain concern was found; the diff changes local TypeScript OS metadata helpers and tests only.

AGENTS.md: found and applied where relevant.

What I checked:

• Repository policy read: Root AGENTS.md and the scoped agent runner AGENTS.md files were read fully; the review applied their whole-path review, focused helper-test, and real behavior proof guidance. (AGENTS.md:1, d1cbe29f3d52)
• Current main still has the reported runtime prompt path: The CLI, embedded attempt, and compaction runtime prompt producers still pass os.type() plus os.release() into runtime metadata, so current main can expose Darwin 25.x on Tahoe. (src/agents/embedded-agent-runner/run/attempt.ts:1936, d1cbe29f3d52)
• Prompt renderer contract: buildRuntimeLine() appends the separate architecture field after runtimeInfo.os, so the correct helper shape is an OS name/version string without its own architecture suffix. (src/agents/system-prompt.ts:1404, d1cbe29f3d52)
• Shipped behavior check: The latest release tag still contains the same raw runtime OS expression in the embedded and CLI prompt producers, so the bug is not already shipped-fixed. (src/agents/cli-runner/helpers.ts:160, c645ec4555c0)
• This PR overlaps the canonical sibling: The PR diff updates the same three runtime prompt producers and adds an OS helper for the same linked Tahoe/Darwin mismatch. (src/infra/os-summary.ts:22, b2f222044bdd)
• Contributor proof remains mock-only: The PR body reports Windows-hosted mocked Darwin tests and explicitly says no live macOS 26/Tahoe runtime-line proof was available. (b2f222044bdd)

Likely related people:

• vincentkoc: Recent GitHub path history shows work in the CLI helper and agent runtime files, and local blame in this shallow checkout points the current raw runtime OS lines at the shallow boundary commit authored by Vincent Koc. (role: recent area contributor; confidence: medium; commits: 92d1f04de340, 0ea08076c3b5, fa4f1abb290c; files: src/agents/cli-runner/helpers.ts, src/agents/embedded-agent-runner/run/attempt.ts, src/infra/os-summary.ts)
• steipete: GitHub path history shows repeated status, OS-summary, and system-prompt work around this metadata boundary, including the status output work that introduced the existing macOS product-version precedent. (role: adjacent feature-history owner; confidence: medium; commits: 1eb50ffac476, a9a386dee19f, 1d7d8a1658b3; files: src/infra/os-summary.ts, src/infra/system-presence.ts, src/agents/system-prompt.ts)
• zhangguiping-xydt: Recent history shows runtime prompt metadata work adding session identity across embedded and CLI-backed prompts, adjacent to the affected OS metadata fields. (role: recent runtime prompt contributor; confidence: low; commits: 78c66742ab97; files: src/agents/system-prompt.ts, src/agents/cli-runner/helpers.ts, src/agents/embedded-agent-runner/run/attempt.ts)

Codex review notes: model internal, reasoning high; reviewed against d1cbe29f3d52.