Document v0.8 command intelligence scope and physical acceptance QA

[raiseCatError]

The v0.8 review stack now has a durable architecture/scope record, ROADMAP next direction, and one deduplicated physical terminal QA checklist. Documentation distinguishes implemented native behavior from external completion research, optional adapters, measured limitations and intentionally deferred scope.

Dependency: #240; merge last, after #193 → #233 → #234 → #235 → #236 → #237 → #238 → #239 → #240. Refs #140, #141, #142, #143, #144, #145, #147, #152.

All code heads have successful Node 22/26 CI; the final code head passed 714 local tests, build, typecheck, benchmark types and git diff --check. Documentation local links and diff checks pass. Cumulative CI is dispatched for this acceptance head too. Physical QA has not been performed. Package version remains 0.7.0; nothing is merged, tagged or released, and issues remain open until release.

Review docs/design/command-intelligence.md and perform docs/qa/v0.8.0-physical-qa.md once after the stack is reviewed/integrated. #52 and all unrelated backlog remain parked. No new issues were created. Existing milestone #7 was renamed to an unscheduled direction as authorized; milestone #8 holds the eight requested command-intelligence issues. Unrelated PRs #181/#182/#183/#226/#227 are untouched.

[raiseCatError]

Final review handoff

Started from fetched/current dev 2d0e8f904896af2947d109a42df85b097b8ebc9a; GitHub dev still has that SHA. Milestone #8 contains exactly #140, #141, #142, #143, #144, #145, #147 and #152, all open. No proposed issue was excluded. As authorized, #7 now names an unscheduled Discoverability & Integrations direction and retains its unrelated contents. #155 remains a prerequisite investigation outside milestone #8.

#193 was salvaged on its existing published branch by merging current dev and updating benchmark coverage, without rewriting/force-pushing history. Nine focused PRs were added. Exact stack/merge order follows the table; each current base is the preceding branch (first base dev).

PR	Branch	Head SHA	Node 22/26 CI
#193	tooling/155-performance-benchmarks	744109e1edab89dac807ce40d4a0385b1341c9bf	Passed
#233	feature/140-structured-completion	89b450d35721b637e9c732e5fd9c07d7c10e9d63	Passed
#234	feature/141-completion-menu	cd5743291f2ac0979066e2e233ccd48c49f8071f	Passed
#235	feature/143-command-history	526f6bb8ffbfb41bf081ee7f2759c5ce6a2af5dc	Passed
#236	feature/144-explicit-history-provider	9c84118c322e061c5f7c2ea4a19ed8c80c49c211	Passed
#237	feature/145-picker-providers	92240b3f5c5225fe56857e9a65baef510d88636b	Passed
#238	feature/147-directory-navigation	21c0e83f19ba9fca1a567672952348327bac030c	Passed
#239	feature/152-safe-command-corrections	8b8e19758427cede31b66f5120031394ac001261	Passed
#240	fix/v08-intelligence-hardening	ff0fa1cc0605bf2be9d66112ed6137fd7baea51c	Passed
#241	docs/v08-command-intelligence-acceptance	2da4d025ff0a352a90e7594c7dc1083d6c5e0fe5	Passed

All local implementation layers passed build, typecheck, full tests and diff checks. Final cumulative local suite: 714 passed; benchmark-script types and documentation local links pass. Navigation CI's initial Node-22 GNU screen cleanup ENOTEMPTY failure and successful unchanged-head rerun are recorded on #238. The status viewport/no-color failures and inconclusive empty RTK log were recorded and resolved with focused/raw verification rather than hidden retries.

Implemented architecture:

• Source-independent structured native completion; bounded detached capture, context cache/fuzzy filtering, generation cancellation and pre-insertion stale checks. Shared actions/glyphs/Chroma and ScreenPlan own presentation.
• Native command-history index derived from eligible additive journal metadata and imported zsh history; worker journal projection and chunked query/import work. No command database duplication/native binding. Independent hashed tombstones preserve concurrent deletions; transcripts remain intact, unknown old journals excluded.
• Explicit Native-default history/picker/navigation settings. Optional Atuin reads a bounded local formatted list, never records/deletes/syncs; metadata duration is approximate and fallback is truthful.
• Supplied-candidate Native/fzf/Television boundary with validated selections and finally-restored host modes. Resize cancels. Installed fzf transport was tested; Television is absent locally.
• /dirs and palette rank native cwd frecency; zoxide queries a temporary database copy because upstream query otherwise sorts/saves. Real fixture source-byte preservation tested. Selection inserts literal cd; separate Enter uses normal zsh/transcript submission. Ordinary cd unchanged.
• Conservative exit-127 correction requires diagnostic, simple syntax and one executable edit. Tab edits only; typing cancels; unsafe/ambiguous input suppressed. UI remains outside output/copy/history.

Full architecture, implemented/deferred scope and benchmark methodology/results are in the acceptance branch.

Representative macOS arm64 / Node 26.8.1 p50/p95 ms: completion filter500 0.06/0.13; editor 8.11/9.13; structured query100k 11.38/11.70; navigation rank100k 6.56/9.26; cached navigation 0.02/0.03; history index import100k 107.08/121.41 total with regular yields. Cold native capture and full100k transcript wrap/JSON remain above warm per-keystroke budgets; no end-to-end socket/startup optimization claim.

External completion research is delivered on #142: Carapace JSON is a promising transport but callback execution lacks an established universal safe boundary; Fig-compatible modules/generators require audited inert-data extraction and per-spec licensing; Inshellisense parser/cache patterns are useful but its execution/UI/runtime constraints rule out direct embedding here. No code copied, no external completion provider/follow-up issue added. #144 contains Atuin findings. No new issues created.

Limits: legacy whitespace replacement for complex quoted completion; frontend PATH snapshots for corrections, without live aliases/functions or subcommand/flag guessing; approximate Atuin durations and possible source duplicates; external history supplies100 newest matches; directory-specific frecency ghost text deferred; broader transcript/startup/storage performance work remains #155. #52 and all explicitly parked work stay parked. Unrelated #181/#182/#183/#226/#227 untouched. Project updates unavailable because token lacks read:project; issue/PR activity records work instead.

One physical-QA checklist covers completion, filters/privacy/deletion, Native/installed external pickers, navigation, correction, Ghostty/Terminal.app, Flow/Chat, passthrough and detach/reattach. It has not been performed.

Next: review in table order. Merge #193 to dev, then retarget each child to dev after its predecessor is integrated, inspect its diff/CI and merge; preserve ancestry with merge commits for this stack and retarget children before deleting predecessor branches. Run the single physical-QA pass afterward and report failures against open issues. Keep issues open until release; package remains0.7.0. No merge, tag or release has been performed. Tracked worktree is clean; pre-existing untracked .serena/ remains outside commits.