docs: Prepare TypeScript 0.4.0 handoff guidance and staged compatibility rollout #31

Scope and current approval

The user explicitly approved reusing this issue for TS 0.4 documentation candidate and Draft PR preparation/validation only on 2026-10-17. No merge, tag, skill/runtime publication, user-installed package update, operations change or live/native transmission/ACK test.

This issue stays separate from the completed public TS skill0.2.1 release (#34/#35) and Python canonical guidance. No duplicate open 0.4 documentation PR was found.

Fixed evidence and published baseline

• Companion base: 0f494cabd05901c03161dc91792dcf89f8e96631; published TS skill0.2.1 and its immutable source/hash remain unchanged.
• Public runtime checkpoint: TS0.3.6, source 7ddded836ef79a7e3384a342ca7011c7abbe127c.
• Source-only feature checkpoint: #109 / 55daf83bd8ebe1cd6e67bb98259130213d1cac71. It reports compiled VERSION0.3.5 despite containing unshipped 0.4 work; version/help alone cannot identify an approved 0.4 release.
• Dependencies #106 → #107 → #108 → #109 and TS #69–#72 require separate integration/review. Final 0.4 source/CLI/guard is not established by this fixed candidate.

Candidate coverage

• Explicit private POSIX ledger init/prepare/status/wait/confirm and top-level ack boundaries.
• Exact native Codex0.160.1/0.162.1 on macOS arm64 with original writer/home/storage qualification and warm metadata store; local/qualified POSIX SSH wait.
• Queue/submission, client-ID injection, turn status and operator_confirmed receipt are distinct; no automatic ACK or post-attempt fallback/resend.
• SQL may internally evaluate body-bearing cells; only bounded IDs/status cross to JavaScript/IPC/output. No blanket body-byte non-access promise.
• Wait writes its own ledger state without resubmission. Its observation qualification must come from the original send. Default30s/max60s total, including5s cleanup, applies to opt-in send/independent wait, not every command.
• Linux native observation #130, Windows observation/private ACL #131, voluntary producer/authenticated collector #132 remain future work. No full Python equivalence, Side Session/steering/Relay/MCP/general ACK/wait expansion.

Proposal files and open contract

A complete review-only successor file is deliberately not named SKILL.md, preserving two-skill discovery and canonical published bytes. Proposed values are skill0.3.0 / minimum0.1.0 / full0.4.0 / typescript / probe-help; these are unagreed. Minimum is the ordinary CLI baseline; full is planned feature coverage, not doctor acceptance/security/platform guarantees.

The fixture manifest separates public0.3.x rejection from desired-but-unimplemented0.4 cases. A proposed0.4 row for existing profiles does not automatically accept a new tuple. Exact numbers, finite matrix, negative/error envelopes, complete final bytes and final integrated source require agreement before implementation/publication.

Order: agree content/profile/fixtures → separately review supporting guard → actual runtime publication/source/artifact verification → separately approve/finalize/publish skill → adopt its same published immutable pin in four-language runtime guidance/PARITY.

Acceptance checklist

• Preserve canonical Python and TS0.2.1 bytes/pin and all existing locales.
• Inspect fixed source and prepare user-action-focused candidate guide/full file/fixture proposal.
• Read-only CLI/ledger and observer/privacy cross-check; correct observed wording findings.
• Validate metadata/local links/hash integrity, two-skill discovery, isolated canonical lifecycle and hash-drift refusal.
• Draft PR37, exactheadb8ff909ffc5fe1e5b22c66a5c302be8332a76038: CI2/2 passed; independent static content findings corrected/rechecked. GitHub automatic Code/Security review is not claimed complete while Draft.
• Exact eb7a1a9 canonical content/profile acceptance received and recorded; not inferred from queued communication.
• Actual supporting TS0.4 runtime publication and exact source/archive evidence verified.
• Renewed canonical exact-content agreement, explicit human merge/release approval and actual skill0.3.0 publication completed.
• External TS runtime four-language guidance/PARITY adoption of published af1addb source; handed off once, not yet claimed complete.

Historical source and native acceptance records belong to their recorded TS checkpoints, not a new whole-release qualification. Previous inactive delivery attempts were not queue consumption; the current user-authorized collaboration will send one exact candidate handoff after preparation.

2026-10-18 candidate update

Draft37 head38cc7be66eb596f9c66df1f7c52ff00e190f6653, fullcandidate22522bytes/SHAbbdcaa4db4245d24127a5c6fefed801ddc7082298278ddd81d5a95fc6e8a0ce9. Metadata proposal0.3.0/min0.1.0/full0.4.0 remains unchanged; earlier direction acceptance does not freeze new full bytes. Merged106/107 security/continuity checkpoints are layered separately from historical109 and still-open108. AES-GCM digest-only journal/separate private key/fail-closed history, Claude private continuity/publicgenerationnull, original waitoperation and duplicateconfirmation invariance documented. Candidate0.3.6reader source verified;0.4reader/runtime defaults remain unfinalized.

ExactheadCI2/2 passed; source/static review only, no actual0.4runtime/profile fixture execution or live experiment. Public Python/TS canonical bytes/pins unchanged. Draft kept; exact final source/profile/full-byte agreement, actual runtime publication and separate skill approval/pin adoption remain open.

2026-10-18 integrated-source candidate

Draft37 new heada969260cd511b7b81631884ff856edaf0d0d457b, completecandidate22973bytes/SHA7571c79fb7ac164002daf0bf3aa7c5ba40ca8d99f5610a04d7860eee7b27df68. Primary source is now unpublished PR151bfbadfa42513791899fc8d2b5441238bbe5b2c4e compiled0.4.0, based on integrated mainbe6adb26f844e7158362a8fcd6d379fbe8d46864. #108/#109 review→squash history verified; aa4/55daf moved to historical-only scope. Queue process/build/executable/probe-budget/post-fence-owner/FD binding refinements and their non-guarantees documented.

Earlier38cc content/profile direction and reviewer-reported candidate validation remain historical, not acceptance of these changed full bytes. The finite0.4 tuple row exists in source but npm0.4 is not published. Actual release/source/artifact verification and separately approved skill finalization/publication/pin adoption remain open; canonical Python/TS0.2.1 and current published pin unchanged.

Runtime publication and canonical finalization — 2026-10-18

Supporting TS0.4.0 is actually published at dde3923efe116f87f38b1639b29b889d2b832e44. Public tarball SHA84adc8b50ef8f03a2c2170e3c36a97b573f28a6bcbeb7dbc965e16c3b37a57c9 and registry/SLSA source bindings checked; tree matches reviewedbfb.

Draft37 head eb7a1a9 prepares canonical skill0.3.0/min0.1.0/full0.4.0 after that prerequisite. New canonical22,839bytes/SHA c510bcfee1d4d1f9c3a992d7ad25d69c9a9881950c627eec60f3b45659071b9e requires renewed exact-content acceptance and separate human merge/release approval. Earlier a969 candidate remains byte-identical history. Python canonical and actual existing0.2.1tag/pin are not changed. Operational guide renamed typescript-handoff.md; installation requires an approved published immutable skill/source/hash and never guesses a pin.

No skill publication/tag/merge/user-installation/operations/native experiment performed. Current finalization checks and forthcoming current-head CI are separate from earlier proposal tests.