RC: DSHX Production 1.0 - #11

Historical note (superseded): This merged PR records the earlier v1.0.0-rc.8 local-IPC architecture. Its bridge/UDS topology and exact-artifact evidence are retired and must not be reused for current promotion. The active product architecture and release evidence are tracked by PR #17 and issues #12–#15: official dsh --profile tui is the sole runtime owner, and the mounted presentation row communicates with the pinned Codex TUI over anonymous inherited directional pipes (fd 0 / 3 / 4), with no bridge executable or socket listener.

Product goal

Ship dshx as a daily-use, production-quality Codex-grade TUI for official DeepSeek Harness. Users run dshx in a project directory; DSHX owns only the pinned Codex TUI thin fork, launcher, packaging, local presentation transport and thin public-API/event projection. DeepSeek Harness remains authoritative for Agent, Session, model routing, tools, sandbox/permissions, approvals, persistence, skills, fork, compaction and subagents.

Release topology (historical; superseded by PR #17)

dshx
  -> official DSH composition (Node, in process)
  -> JSONL presentation relay over child stdio
  -> packaged dshx-ipc-bridge
  -> private cross-platform Unix-domain socket
  -> pinned Codex TUI thin fork

The topology above is retained only as history for the already-merged rc.8-era work. Current production has no bridge executable, UDS/WebSocket listener, or TUI-owned backend child.

Implemented daily-use surface

• zero-argument dshx launch and dshx resume;
• streaming assistant + reasoning;
• DSH-backed model picker and persisted default model;
• DSH permission presets and fail-closed approval UI;
• Ask User;
• shell/tool/diff presentation using DSH tool presenters/execution;
• plan mode/todos;
• steering and Ctrl+C interruption;
• session list/resume/paginated history/token-context footer;
• durable DSH session rename, fork and manual compaction;
• skills;
• subagent discovery/status with DSH-owned authority;
• isolated DSHX presentation home (never the user's normal CODEX_HOME);
• unsupported Codex-only runtime surfaces hidden rather than shadow-implemented.

1.0 SWAT workstreams

• 1.0 SWAT: Freeze dependency graph and switch all gates to npm ci #13 — Release Engineering: frozen dependency graph, npm ci, reproducible release closure.
• 1.0 SWAT: Cross-platform CI, package, and exact-head evidence #14 — CI / Platform: exact-head cross-platform build, Linux/macOS PTY + Windows ConPTY TUI smoke, package, clean-install, tag evidence.
• 1.0 SWAT: Runtime ownership and directional-pipe security audit #15 — Runtime / Security: DSH ownership and fail-closed semantics for the current directional-pipe architecture.
• 1.0 RC gate: Windows Terminal / CJK / IME side-by-side parity #12 — UX acceptance: exact Windows RC artifact in Windows Terminal, Chinese IME/wide-glyph/mouse side-by-side parity.

The active integration/promotion surface is PR #17; the workstream issues are the release evidence ledger.

Release gates

Current gates are defined by the repository at the exact candidate SHA and docs/RELEASE-READINESS.md. Do not reuse the rc.8 bridge/UDS evidence for the current candidate.

The final 1.0 promotion additionally requires the manual Windows Terminal side-by-side UX gate, especially Chinese IME composition and wide-glyph alignment, against the exact generated RC artifact.

Review rule

Do not tag final 1.0 merely because checks exist. Require green evidence for the exact head/tag. Any missing DeepSeek Harness capability stays hidden/unsupported; do not solve a release failure by adding a shadow Agent/runtime implementation to this repository.