Spec Kitty 3.2.3 patch — debugger-debbie investigation

Branch fix/3.2.3-coord-surface-regressions @ 775ec32da (off v3.2.2 main), editable install active. Lens: LIVE-confirm each ticket + root-cause; classify split-brain (coord/primary surface-resolution); does SSOT/topology-alignment fix it. All reproductions are real runs (CLI or live Python against the editable install), not static reads. Repo left clean (read-only).

Per-ticket verdict table

#	reproduced (Y/N + repro)	root cause (file:func:line)	split-brain? (Y/N + facet)	SSOT/topology fixes it? (Y/N + seam)
#2122	Y — `pytest tests/specify_cli/test_acceptance_regressions.py::test_collect_feature_summary_anchors_feature_dir_on_primary_for_mid8_handle` FAILS: `AcceptanceError: Feature '01ABCDEF' has no tasks directory at .../kitty-specs/01ABCDEF/tasks`. Real entry point `collect_feature_summary(repo_root, mid8)`.	`acceptance/__init__.py:_iter_work_packages:406` (→ `_wp_tasks_read_dir:827` → `resolve_planning_read_dir`) AND `acceptance/__init__.py:normalize_feature_encoding:645` (→ `_planning_read_dir` → `resolve_planning_read_dir`). Both pass the raw handle to `resolve_planning_read_dir(repo_root, feature, kind=…)` → `primary_feature_dir_for_mission:1212` which joins the handle LITERALLY to `kitty-specs/<handle>/` (line 1240, no handle→slug step).	Adjacent, NOT the stale-coord facet. Facet = handle-resolution-seam asymmetry: the gate has a handle-AWARE anchor (`_primary_anchor_feature_dir:889` → `resolve_mission`) and a handle-BLIND read (`resolve_planning_read_dir`); two call sites use the blind one.	Y — fix = resolve handle→canonical slug FIRST (via the same `resolve_mission`/`_primary_anchor_feature_dir` seam the summary already uses), then pass the slug to `resolve_planning_read_dir`. Restores the handle-awareness `resolve_feature_dir_for_mission` had pre-#2113.
#2120	Y — built a real coord-topology mission (materialized `-coord` worktree, coord branch, lanes.json); ran `spec-kitty mission close --mission demo-coord-feat --discard --force`. Output: `✓ Mission … discarded.` yet `git worktree list` and `git branch` BOTH still show the coord worktree + branch.	`cli/commands/mission_type.py:close_cmd:595` resolves via `resolve_feature_dir_for_mission` (coord-aware) → returns the materialized coord dir (no meta.json, live-proved `meta? False` for all handle forms). Then `_read_mission_mid8:632` returns `""`, so `_teardown_coordination_worktree:717 (if not mid8_value: return)` early-returns → no teardown. Line 626 prints `✓` unconditionally. Secondary: `_discard_mission:658-659` deletes lane branches BEFORE removing the worktree they're checked out in (`git branch -D` fails, swallowed `check=False`).	Y — THE canonical split-brain facet. `close --discard` resolves coord-aware; `reopen` resolves primary (`resolve_mission`/`primary_feature_dir_for_mission`) → two commands, same mission, different surfaces (DIRECTIVE_032 conceptual-alignment fork).	Y — LIVE-PROVED: resolving the discard path via `primary_feature_dir_for_mission` yields `meta? True`, `mid8='01ABCDEF'`, and `CoordinationWorkspace.teardown` then succeeds (`WORKTREE GONE`). Seam = the same primary-anchor `reopen` already uses (ticket option A). Add ordering fix (B) + non-zero exit on incomplete teardown (C).
#2119	Y — (a) with a materialized coord worktree, `canonical_record_path(repo, handle)` resolves the retrospective WRITE target to `.worktrees/<slug>-<mid8>-coord/kitty-specs/<slug>-<mid8>/retrospective.yaml` (live-printed, `IS on coord worktree? True`) — on the ephemeral coord branch. (b) `spec-kitty agent worktree --help` → `No such command 'worktree'` (live).	(1) `retrospective/writer.py:canonical_record_path:36-49` resolves via `resolve_feature_dir_for_slug` (coord-topology-aware) → durable home lands on the ephemeral coord branch, never `.kittify/missions/<mission_id>/`. (2) Dead-end remediation text in `coordination/surface_resolver.py:119` (`_COORD_EMPTY_FALLBACK_WARNING`) and `:203` (`CoordinationBranchDeleted.next_step`) + `cli/commands/doctor.py:3197/3221` all recommend the nonexistent `spec-kitty agent worktree repair`.	Y (facet 1) — same write/read-resolved-through-coord-seam class as #2120, plus a teardown-vs-retro conflict (only copy lives in the worktree merge/close removes) and a dead-end-command facet (orthogonal docs bug).	Y (facet 1) — anchor the retrospective DURABLE home on PRIMARY (e.g. `primary_feature_dir_for_mission`, or the `.kittify/missions/<mission_id>/` the mission-review skill already expects), so merge teardown neither blocks on nor destroys it; flatten meta on merge. Partial for the rest: (3) make `retrospect create` tolerate a flattened/torn-down topology (don't hard-block on `CoordinationBranchDeleted` for an already-merged mission); (4) fix the remediation text — implement `worktree repair` or point at the real flatten path.
#2112	Y — `spec-kitty init <name>` creates SUBDIR `<name>/.kittify/config.yaml`, never `cd`s in and never `git init`. `spec-kitty specify` then errors `SPEC_KITTY_REPO_NOT_INITIALIZED … Missing: …/.kittify/config.yaml` — byte-matching the report (two modes: no `.git` anywhere → resolved-root=project but reported Missing; OR a parent `.git` exists → resolver walks UP to the parent, resolved-root=parent). `/spec-kitty` in Cursor works because the slash surface runs the agent prompt/scripts, not the Typer guard.	`core/paths.py:resolve_canonical_root:417-419` returns the first ancestor whose `.git` is a directory with no `.kittify` boundary check (walks past the project's own `.kittify`); and `:437` raises `WorkspaceRootNotFound` when no git marker exists even with a valid `.kittify/config.yaml`. Surfaced via `workspace/assert_initialized.py:assert_initialized:92-105` ← `cli/commands/lifecycle.py:_enforce_initialized:100-101`. The sibling `locate_project_root:254-265` DOES have the `.kittify` fallback — classic two-parser divergence.	N — repo-root detection, NOT coord/primary split-brain. No worktree/coord branch/kitty-specs involved; a plain ancestor-walk landing on the wrong root for a new single-checkout project. priti's triage CONFIRMED.	N (not the coord SSOT). Fix is upstream of topology: give `resolve_canonical_root` a `.kittify`-marker boundary/fallback so it agrees with `locate_project_root` (a root-authority dedup). Dedup: NOT a clean dup — it is the un-fixed residual of #2011 (same function; #2011's patch fixed the submodule pointer branch but left the `.git`-directory + no-git branches). #539 is a DIFFERENT bug (worktree-root misroute). Link to #2011, extend that fix; do not close as dup of #539.
#2116	N/A (tech-debt, not a live bug) — `wc -l tasks.py` = 3365 LOC (vs ~1200 target); `ruff … --select C901` = "All checks passed" (maxCC≤15 IS met).	(a) body-thinning only — mega-commands `move_task`/`mark_status`/`map_requirements` undecomposed (`agent/tasks.py`, `move_task` def L899 ~770 LOC). (b) `tasks.py:_skip_target_branch_commit:530` + `_protected_branch_status_commit_error:497` govern a load-bearing exit-0 silent-skip + `--json` envelope reshape (`status_events_path`→coord, `wp_file_update:"skipped"`, L1622-1644) that `commit_for_mission` cannot reproduce (its protected-primary path returns `no_op_wrong_surface`, an error-class refusal). (c) pre-existing fork: `move_task` skips exit-0 (L995/L1024) but `mark_status:1798` / `map_requirements:2425` REFUSE exit-1.	N — coord-ADJACENT exit-semantics, NOT the stale-surface-read class. Here surface resolution is CORRECT; topology is correctly detected and the commit is skipped/refused. The defect is inconsistent exit semantics across sibling commands (DIRECTIVE_032 fork) — no stale read, no data loss.	Partial. (a) orthogonal (pure decomposition). (b)/(c) addressable by teaching the router ONE topology-aware protected+coord exit policy (a router-contract change affecting all callers — exactly what #2114 deferred to stay behavior-neutral).

Cross-cutting structural read (DIRECTIVE_001 / 032 — divergence matrix)

The single recurring structural fault across #2120, #2119(facet 1), and #2122 is one boundary violation: a primary-anchored operation is routed through a coordination-topology-aware resolver, so it lands on the ephemeral -coord surface instead of the durable primary checkout.

Operation	Wrong (coord-aware) seam used	Correct (SSOT-aligned) seam	Symptom
`close --discard` identity/mid8 read	`resolve_feature_dir_for_mission` (mission_type.py:595)	`primary_feature_dir_for_mission` / `resolve_mission` (as `reopen` does)	silent no-op teardown, `✓` lie
retrospective DURABLE write	`resolve_feature_dir_for_slug` (writer.py:48)	primary anchor / `.kittify/missions/<id>/`	retro destroyed/blocked by teardown
accept-gate WP-task + encoding read	handle passed blind to `resolve_planning_read_dir` (acceptance:406/645)	resolve handle→slug FIRST, then primary read	`--mission <mid8>` accept-gate crash / silent no-op

primary_feature_dir_for_mission is topology-blind and correct but handle-blind (literal join, no mid8→slug). So the unified fix has TWO obligations:

Anchor primary-partition operations on the primary surface (#2120, #2119-write) — stop using the coord-aware resolver for durable/identity reads.
Resolve handle→canonical slug at the input boundary before the literal primary join (#2122) — the gate already owns this seam (_primary_anchor_feature_dir/resolve_mission); the two stragglers must reuse it, not the blind primitive.

These are the same single-authority discipline the surface-resolver missions (01KVN754 / #2040) established; these three are residual non-adoption sites.

Coherent 3.2.3 SSOT/topology-alignment slice

Tier 1 — same-class split-brain (one coherent slice, ships together):
- #2120 — discard path resolves PRIMARY (option A) + worktree-before-branch ordering (B) + non-zero exit on incomplete teardown (C). Red-first: the deterministic close-discard coord regression.
- #2119 facet 1 — retrospective durable home anchored on PRIMARY (+ flatten meta on merge). Red-first: assert canonical_record_path for a materialized-coord mission is NOT under .worktrees/…-coord/.
- #2122 — resolve handle→slug before resolve_planning_read_dir at _iter_work_packages + normalize_feature_encoding (and audit any other gate-read site for the same blind-handle bug). Red-first guard already exists (…_for_mid8_handle).
Tier 2 — coord-ADJACENT, fold in but separable:
- #2119 facets 3/4 — retrospect create tolerate flattened topology; replace the phantom agent worktree repair text (surface_resolver.py:119/203, doctor.py:3197/3221) with the real flatten path (or implement the command).
- #2116 (b)/(c) — unify the protected+coord exit semantics into ONE router-owned policy (topology-aware exit, not stale-read); (a) body-thinning is orthogonal cleanup.
NOT in this slice — different root authority:
- #2112 — repo-root detection (resolve_canonical_root .kittify boundary/fallback). Residual of #2011, NOT coord split-brain. Fix independently; link to #2011 (not a dup of #539).