[abruption]

Summary

Adds the first, deliberately limited slice of #323: session-peer send --dry-run --explain describes existing ordinary local/SSH dry-run evidence in human-readable text, separately for each SSH destination.

session-peer send --host worker --to 'codex:<full-thread-uuid>' --message "Review the patch" --dry-run --explain

• Reports the requested/resolved destination, reported agent home, planned delivery method, existing validation/wake preflight and observation limits.
• Distinguishes unknown outcomes and unexpected reported submission evidence from a validated preview. A common input/option failure is not presented as a destination check; a later host's failure is not attributed to earlier successful hosts.
• Does not add probes, forward --explain over SSH, create placeholder messages, change native optional fields or alter ordinary JSON/wire/exit behavior. New display fields are allowlisted and terminal-sanitized.
• Rejects missing --dry-run, JSON output, paired-device and experimental handoff combinations before update preparation, input reading or destination/ledger work.
• Updates the English, Korean, Japanese and Simplified Chinese CLI references and their source manifest; regenerates the standalone entrypoint.

Refs #323. This does not complete the issue's broader paired-device, handoff or discovery presentation scope.

Type

• feat
• docs
• tests

Validation

• macOS arm64, Python 3.13.7 with relay/MCP test extras: 1341 tests OK, 19 skipped, with development mode and ResourceWarning treated as an error, at final head 3302f66e185352e68e4c22a0519e28b0e6713f4c.
• Focused compatibility suite before the final fixture correction: 95 tests OK, 1 skipped on Python 3.9.6 and 3.13.7. After correction, all 22 explanation-specific tests pass on both versions: the Windows-refusal negative explicitly exercises the platform branch on every runner, while the real file-success case remains POSIX-only.
• Four-language documentation/manifest checks are included in the full suite; standalone generator --check and git diff --check pass.
• Independent read-only CLI reviews were run before PR creation. Confirmed findings about uncertainty, per-host attribution and missing dry-run evidence were corrected and regression-tested. These are static reviews, not native transport acceptance.

Compatibility and limits

• Existing output without --explain, machine-readable fields and exit codes are preserved, including legacy Claude field absence and the existing stdout/stderr placement of old summaries/errors. A new failure or contradictory-evidence explanation uses stderr.
• A preview is not a reservation or proof of later submission, consumption, receipt ACK or task completion. A later send repeats existing validation. Unknown outcomes never authorize automatic resend.
• Wake is not checked unless explicitly requested; only existing preflight evidence is described. Ordinary receipt ACK support is not introduced.
• Tests are bounded offline fixtures with native submission and SSH execution mocked; no real agent messages, new native/platform acceptance, Relay/MCP operations, global installation or deployment was performed.
• No version bump, merge, tag, release or package publication is part of this PR.

CI follow-up

• The first Windows run exposed a new test's invalid assumption that POSIX-private message-file input succeeds on Windows. The product already refuses that input on Windows. The follow-up changes tests only: the positive case is POSIX-only, and a Windows-specific negative case verifies the existing refusal before destination checks or native calls. Product behavior is unchanged.
• Separate blocker: Control dependency auditing reports the existing locked vitest / @vitest/mocker 4.1.11 and vite 7.3.6 advisories. Control code, dependency files and workflows are unchanged by this PR. This is not a fully green CI claim, and no dependency upgrade, audit bypass or merge is included here.
• Latest-head Windows Python 3.13 CI now passes. Automatic Codex code and security reviews completed with no findings at product-code head 03e9891; the later two commits modify tests only. Do not treat that as a new automatic review of the later test-only head. Remaining CI jobs may still be running/queued.

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-22T07:15:17.679687Z	03e9891	PR opened
🔒 Security Review	✅ Completed 2026-10-22T07:16:11.582818Z	03e9891	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.

[abruption]

The unchanged Control dependency-audit blocker is now tracked separately in #326 and fixed by dependency-only PR #327. That PR updates compatible Vite/Vitest patches, preserves the existing audit gates, and has successful local production/full audits. After normal approval and merge of #327, refresh or rerun this PR against the updated main; this comment does not claim current #325 CI is green.