[abruption]

Summary

Refs #186; intentionally does not complete the issue.

• Compare a Python adapter, independent TypeScript client, and no npm distribution. Select a bounded Python-free, read-only discovery prototype, not a full messaging port.
• Add experiments/node-cli, excluded from stable Python distributions. Explicit local Claude/Codex inventory only; Codex requires --codex-home.
• Fail explicitly for send (including dry-run), implicit inventory, SSH, paired devices, Relay client/server, MCP, wake and unsupported options/platforms. No external process or network invocation in the runtime.
• Add Python-v1 cross-runtime conformance, native SQLite/Unix socket fixtures, package-content/repeat-pack/clean-install/uninstall checks, and macOS/Linux Node 22/24 CI.
• Keep package private, prototype version 0.0.0, reference contract 1.0.2, and a nonconflicting executable name. No separate repository or npm publication.

Type

• feat
• chore / refactor / test / perf

Checklist

• Commits follow Conventional Commits
• Prototype scope, prerequisites, differences and release gates documented
• No secrets, tokens, or session keys in the diff
• Existing Python runtime, generated standalone, Relay and deployed installations unchanged
• Real SSH / send / dry-run / native inbox / queue / ACK interoperability: deliberately not supported or claimed in this slice

Verified against

• macOS arm64, Node 22.14.0: 8 conformance tests pass against the repository's Python CLI and compatibility-v1 fixture.
• Actual temporary Unix socket (metadata only, zero connections), real temporary SQLite databases; no real user sessions/messages.
• Fresh npm-prefix installation, empty-PATH execution, uninstall and repeated tarball SHA-256 equality pass.
• npm audit: 0 vulnerabilities, including build dependencies.
• Python compatibility module: 3 pass / 1 optional-relay-dependency skip.
• Generated standalone check and git diff --check pass.
• Node 22/24 macOS/Linux matrix runs in the new workflow; local results do not substitute for those CI results.

Notes

This is a draft for the first #186 gate. The remaining work is explicit in the experiment README: full inventory, live-writer kernel lock/process identity validation, pre-submit revalidation, envelope/reply and uncertain outcome contracts, authorized real inbox/queue tests, SSH runtime decision, platform/release/provenance gates.

Saved Codex rows and Claude reachable metadata are not evidence of consumption or ACK. Unsupported commands return submitted=false and never fall back to Python. Error messages deliberately use fixed codes rather than echoing untrusted arguments or SQLite contents. Read-only SQLite is not a snapshot and may participate in WAL shared-memory bookkeeping.

[abruption]

Superseded by the user-approved repository split: https://github.com/abruption/session-peer-ts (initial split head 326861cf9ac5b05be0c14a81424919bd9b0b35ce). The independent CLI now implements guarded local and SSH delivery. Four fresh one-shot ACKs were independently observed in dedicated native Claude/Codex TUIs across local Mac mini and Mac mini → MacBook SSH; see VALIDATION.md in the new repository. This draft is not merged into the Python runtime. Issue #186 remains open for the remaining public npm release gates; no npm publication occurred.