[abruption]

Summary

Completes the approved v1.3 explanation scope for #323, expanding the original local/SSH text dry-run slice:

• send --dry-run --explain covers local, SSH and paired receivers; doctor --explain explains the existing diagnostic snapshot; handoff status --explain explains an existing read-only intent snapshot.
• Text only. JSON combinations and handoff send combinations are refused before advisory/input/destination work. Ordinary output, native optional fields, exits and the frozen handoff schema/fixture remain unchanged when the option is absent.
• Distinguishes declared capability, this invocation's validated facts, unavailable/unknown evidence, wake preflight, receipt assurance, latest wait and task completion. A receipt never proves task completion or model identity. Explanation generation time is not delivery-observation time; receipt time is labelled as stored record time.
• Paired explanation context is negotiated through the existing authenticated probe, only for an explicit resolve. A new receiver advertises delivery-explanation-v1; an old receiver gets the unchanged resolve body and preserves known native facts with unknown context. No extra probe, connection, native submission, collector or fallback is introduced.
• The private context is a strict, bounded 8 KiB v1 envelope. It uses only executing-worker declarations and actually returned native target/home. Unsolicited, malformed or unrequested context is discarded; optional metadata cannot displace otherwise usable native facts at the existing result byte cap. Actual direct/relay is shown only when the existing response reports it.
• Handoff explanations reuse the shared read-only ledger/clock primitives from feat(handoff): add read-only status and bounded request watch #329 without merging that sibling PR. They never persist retention, initialize/repair history, mint authority, create waits or start a collector. A legacy collector's read-only clock rejection does not fall back to a write-through query.
• New explanation text excludes message/URI/error/authority data and escapes terminal/bidi controls without changing ordinary output. Updates the four CLI-reference locales and generated standalone artifact.

Refs #323. No merge, runtime version bump, tag, release, package publication or operational change.

Validation

• Final Python 3.13 extras suite, development mode and ResourceWarning-as-error: 1,373 tests OK, 19 skipped.
• Python 3.9 explanation/CLI modes: 45 tests OK. The new relay path is not claimed for Python 3.9; existing optional relay requires Python 3.11+.
• Documentation/manifest: 65 tests OK; generator check and staged diff checks pass. Frozen handoff contract/fixture unchanged.
• The initial full test ran while documentation was being edited and failed five doc checks; the final frozen run above includes the corrected docs and all review regressions.

Tests use isolated offline fixtures/mocked native operations and loopback components; no real agent/model/SSH-account/hosted deployment acceptance is claimed.

Independent code review

Before push, independent-cli-review ran Claude Code 2.1.308 (requested and observed claude-opus-5-5) and Codex CLI 0.163.0 (isolated default; actual model not exposed) in parallel. Both received identical frozen self-contained source/test excerpts, completed with exit 0, and left input SHA-256 a6e7c8b3f931c111104ff26ac08361b34c391663904ecbb5a9a4393f580c35b1 unchanged. Claude reported no execution error; Codex completed its turn without command-execution events. Reviewers had no file-edit, test, network or external-session authority.

Confirmed findings were corrected: Claude context prefix, paired real-CLI validation-stage attribution, actual-route display, unsolicited context rejection, failure evidence labels, stored receipt time wording and bidi sanitization. Root additionally guarded the optional-context byte-cap boundary. Regressions exercise actual generated code and real CLI-to-renderer paths rather than mocking the formatter. A separate read-only subagent inspected the corrected integration and found no blocker. These corrections are not represented as a second exact-head CLI review or live security/transport acceptance.

Translation quality review

Antigravity CLI 1.4.5 explicitly selected available gemini-3.8-flash-medium, medium effort, plan mode and sandbox, five-minute limit. One fresh structured review completed with exit 0 in 147.064 seconds; no permission bypass, repository exposure, tools, network or edits were authorized. Input covered the frozen changed explanation sections in English/Korean/Japanese/Simplified Chinese. Prompt SHA-256: a6e19a9283239a750f16029835e2c0242ac4f553ee1108901bd4f78effed3caf; evidence SHA-256: fc8d9da6c7e1e2f723da1a519b9d14f36ee96cc126ecb5b7cba823949def4588.

Codex checked proposed edits against the English contract and corrected Japanese context wording and Chinese no-new-schema wording. Overstated model claims about doctor wording were not adopted. Code, options, links, numbers and supported-scope limits are preserved. The selected model is recorded; independent backend identity is not exposed. This is linguistic/semantic QA, not runtime/native validation.

Integration gates

• Existing sibling PRs remain separate and unmerged. Combined-tree verification is required before release, including the shared feat(handoff): add read-only status and bounded request watch #329 reader and feat(discovery): add opt-in local destination presets and address book #324 address workflow.
• Ordinary readiness is a snapshot, not target reservation or atomic storage/owner attestation. Unknown outcomes never authorize automatic resend.
• Companion skill documentation and release preparation remain separate. No source is installed globally and no operating configuration is changed.

[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.

[abruption]

The owner-approved additional #323/#324 implementations are pushed to the existing PRs:

• feat(cli): explain delivery evidence across supported routes #325: 5507298a699cc9462500b830e28120e121de7cbb, all 28 checks pass.
• feat(cli): add checked route presets and saved recipient bindings #328: d004928f7eeb9f9985f23730d90b0bfd80f92ddd, all 28 checks pass.

Independent Claude/Codex code-review findings and verified Gemini translation suggestions were addressed before the review handoff. Precise model/input/version limits are recorded in the PR bodies; this does not imply a new cloud review of an identical final tree.

A separate local combined tree includes all six ordinary v1.3 feature PRs (#325/#328/#329/#330/#331/#332), detached head 0511ef4a6ee7ae0b2866193db963ac18a563e001. Python 3.9 focused composition: 110 pass; documentation: 65 pass; six new cross-feature regressions pass; generator/diff checks pass. Full-discovery audit: 1698 retained pass, 23 skipped, excluding exactly 35 proven external-volume environment-baseline method IDs (29 preset and six original #330 SSH-storage methods). Their failure signatures were independently reproduced; no production guard, test assertion or actual filesystem permission was weakened. This is not an unfiltered combined-tree CI or live delivery claim.

No PR/main merge, version bump, tag, release, publication, installation or operations change was performed. The integration commits are local only; normal sequential merges will still need current-head protections and checks. Research #245/#247 and Draft #246/#248 are assigned to Long-term.