diff --git a/AGENTS.md b/AGENTS.md
index 52b2f1bc..8caca2f2 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -26,17 +26,11 @@ tracks, applied to the instructions. One file, one answer.
 
 ## Writing a design-state record
 
-This repository has adopted **half** of the kit's `design/state/` system, and the half it did
-not adopt is the half this heading is about. What exists: `/agentkit:track`'s work-mirror
-(`design/state/work/`, one `WorkRef` per issue — `design/90-decisions.md`, 2026-08-24) and
-`design/state-index.md`, the projection over it (2026-09-05). What does not: `design/state/units/`,
-and every other record kind. There are no `Unit`, `Invariant`, `Contract`, `Decision` or
-`Question` records here, which is why five of the index's six regions render an empty-set
-placeholder and only `outstanding` carries rows.
-
-So the full record-writing sequence the kit's own `AGENTS.md` describes under this heading —
-appending to `90-decisions.md`, writing a decision record, updating unit records, regenerating
-projections — still does not apply. **Write the decision-log entry alone**, in
-`design/90-decisions.md`'s own register, in its existing format. Do not hand-write
-`design/state-index.md`: it is generated, `/agentkit:track` regenerates it, and an edit between two of
-its markers is discarded on the next run.
+The kit no longer ships `/agentkit:track`, `Test-DesignState.ps1` or `Update-DesignProjection.ps1`,
+so nothing regenerates this repository's `design/state/` records. What remains from the earlier
+half-adoption is the work-mirror (`design/state/work/`, one `WorkRef` per issue — `design/90-decisions.md`,
+2026-08-24) and `design/state-index.md`, the projection over it (2026-09-05); both are frozen as
+they stand until a separate decision retires them (`design/90-decisions.md`, 2026-10-07).
+
+**Write the decision-log entry alone**, in `design/90-decisions.md`'s own register, in its existing
+format. Do not hand-edit `design/state-index.md` or `design/state/`.
diff --git a/CLAUDE.md b/CLAUDE.md
index 18980d91..f7d0173b 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -154,7 +154,7 @@ build context:
   matching the stated reading order. **A new page is invisible until it is listed there**;
   an id that does not resolve fails the build.
 - `build/ConvertTo-HumanDocumentation.ps1` — the design → human-doc generator. Run it after
-  editing marked blocks in `design/`; run `/agentkit:docs` for `docs/docs/guide.md`, then
+  editing marked blocks in `design/`; regenerate `docs/docs/guide.md` by hand, then
   stamp the guide with `./build/ConvertTo-HumanDocumentation.ps1 -StampGuide`. Never edit a
   generated engine page directly. `./build/Test-Documentation.ps1` runs the corresponding
   `-Check` automatically.
@@ -239,7 +239,7 @@ bulk. When a suggestion is declined, record it in the affected document (or
 `OPEN-QUESTIONS.md`) as a known-and-retained issue rather than dropping it silently.
 
 **A reconciliation ends in a decision, not a report.** Any time you compare two things and
-find they disagree — `/agentkit:align`, `/agentkit:install`, `/agentkit:track` drift, a spec checked against the
+find they disagree — `/agentkit:align`, `/agentkit:install`, a spec checked against the
 code, or any time the user says "reconcile" — the work is not finished at the findings.
 Close by asking, one divergence at a time, each with a recommendation and what the
 alternatives cost. **A report the user has to turn into questions is half the job.**
@@ -310,7 +310,7 @@ decides whether a model should be doing it at all.
 
 A red item is a defect in the tooling, not in the run. Noticing one is worth a line; performing it
 repeatedly and never saying so is the failure. When a red item recurs, add it to the open register
-in `design/90-decisions.md` so `/agentkit:track` can turn it into an issue — that is the existing path, and
+in `design/90-decisions.md` so it is picked up as an issue — that is the existing path, and
 there is no separate mechanism for this. This repository has already taken that route once:
 `build/ConvertTo-HumanDocumentation.ps1` exists because regenerating human pages by hand was red.
 
@@ -321,7 +321,7 @@ Two distinctions that are easy to get wrong:
   merge convention governs is not. Do not classify a whole command by its cheapest step.
 - **Do not report a cost you did not measure.** A model is not given its own token counts or
   elapsed time, so any figure it states about its own run is an estimate presented as a
-  measurement. `tools/Measure-Session.ps1` reads the real per-call usage from the session
+  measurement. the kit's `measure-session.ts` reads the real per-call usage from the session
   transcript, and runs as a `SessionEnd` hook. It measures **Claude Code sessions only** —
   Codex writes a different schema this has no reader for, and Copilot records no token usage
   at all. Use it, or say nothing — under either of those, *say nothing* is the whole
@@ -345,35 +345,20 @@ Two distinctions that are easy to get wrong:
 
 ### The Design Freeze
 
-The pipeline's normal loop keeps `design/` live: a unit lands, `/agentkit:align` writes reality
-back, `/agentkit:track` resyncs the tracker. That is right while the design is still being settled
-and **wrong once implementation is the bottleneck**, because each pass is generative rather
-than merely checking — landing W<n> rewrites W<n+1>'s specification, which desyncs the
-tracker, which needs `/agentkit:track`, which finds drift, which needs `/agentkit:align`. The loop has no
-fixed point. Freezing is how it is escaped.
-
-**`design/FROZEN.md` is the marker, and its existence is the whole mechanism.** It is
-tracked, not ignored — a freeze is a statement to everyone working in the repository, not
-local state. While it exists:
-
-- **`/agentkit:align` and `/agentkit:track` do not run.** The tracker is deliberately allowed to go stale.
-- **`/agentkit:design`, `/agentkit:spec` and `/agentkit:plan` refuse.** Authoring is gated too, so the docs
-  cannot drift forward while the implementation is being checked against them.
-- **Units implement against `20-contract.md` as a fixed artifact**, at the SHA the marker
-  names.
-- **A contradiction found while implementing is stated in that unit's pull request and left
-  in the document.** Do not fix it in `design/`. The staleness is the point; recording it in
-  the PR is what makes the eventual reconciliation cheap.
-
-**`/agentkit:hold` writes the marker; `/agentkit:resume` lifts it** — deletes the file, then runs one
-reconciliation pass, `/agentkit:align` then `/agentkit:track`, in the same session. `/agentkit:resume` runs
-unattended, without a confirmation prompt; the freeze itself is still the user's decision,
-made when `/agentkit:hold` is invoked, and lifting it early is one command call away rather than
-gated a second time. A unit that turns out to need a contract amendment still stops and
-says so; that escalation is the user's to answer, and answering it may well be "thaw,
-amend, re-freeze."
-
-The marker's format, which the five gated commands read and must not restate:
+The kit no longer ships `/agentkit:hold`, `/agentkit:resume`, `/agentkit:track` or the gated
+authoring commands, so a freeze is now a **manual convention** with no command enforcing it.
+The reasoning is unchanged: keeping `design/` live while implementation is the bottleneck has
+no fixed point, because each reconciliation pass rewrites the next unit's specification.
+
+**`design/FROZEN.md` is still the marker, and its existence is the whole mechanism.** It is
+tracked, not ignored — a freeze is a statement to everyone working in the repository. While it
+exists, nobody runs `/agentkit:align`, `/agentkit:design` or `/agentkit:plan`, and units
+implement against `20-contract.md` as a fixed artifact at the SHA the marker names. A
+contradiction found while implementing is stated in that unit's pull request and left in the
+document; the staleness is what makes the eventual reconciliation cheap.
+
+The marker is written and lifted **by hand**. Lifting is deleting the file, then running
+`/agentkit:align` in the same session. Its format:
 
 ```markdown
 # design/ is frozen
@@ -382,23 +367,26 @@ Frozen at: <sha>, <YYYY-MM-DD>
 Frozen because: <what the freeze is escaping>
 Lifts when: <the checkable condition — "tier one is code-complete", not "when we are ready">
 
-To lift: run `/agentkit:resume`, or delete this file by hand and run `/agentkit:align`, then `/agentkit:track`.
+To lift: delete this file and run `/agentkit:align`.
 ```
 
-A command that refuses reports `Frozen because` and `Lifts when` **verbatim** rather than
-paraphrasing them — the point of a stated condition is that it can be checked against, and a
-paraphrase is where it stops being checkable.
+Anything that refuses on the marker reports `Frozen because` and `Lifts when` **verbatim**.
+Recorded in `design/90-decisions.md`, 2026-10-07.
 
 ### The Agent Kit — Canonical Workflow
 
-The [agent kit](https://github.com/The-Running-Dev/SubZeroDev.AgentKit)'s pipeline commands
-load machine-wide as the `agentkit` plugin — `/agentkit:<name>`, each read live from the kit
-checkout — so this repository holds no copies of them, only its companions,
-`skills/<name>/SKILL-local.md`. Bare `/plan`, `/help` and `/design` are Claude Code's own; the
-namespace is not optional.
-`design/00-brief.md`, `10-design.md`, `20-contract.md`, `30-slices.md`, and
-`90-decisions.md` are now their real canonical inputs. This repository extends the generic kit
-in two deliberate ways:
+The [agent kit](https://github.com/The-Running-Dev/SubZeroDev.AgentKit)'s commands load
+machine-wide as the `agentkit` plugin — `/agentkit:<name>`, each read live from the kit
+checkout — so this repository holds no copies of them and no per-repository companions. Bare
+`/plan` and `/design` are Claude Code's own; the namespace is not optional. `/agentkit:sync`
+advances the kit; `/agentkit:install` reconciles this repository's own files.
+
+The kit's pipeline is `interview` → `design` → `plan` → `next`, with `fix`, `align` and
+`redteam` alongside, and `install`, `install-all`, `install-review` and `sync` for setup.
+`/agentkit:next` builds the plan slice by slice through merge, in one session. There is no
+separate `slice`, `check`, `pr`, `resolve`, `track`, `docs`, `hold` or `resume`.
+`design/00-brief.md`, `10-design.md`, `20-contract.md`, `30-slices.md`, and `90-decisions.md`
+are the canonical inputs. This repository extends the generic kit in two deliberate ways:
 
 - Existing W identifiers remain the slice ids. Renumbering them to S would break plans, tests,
   issues, changelog entries, and merged history without changing the work.
@@ -411,86 +399,45 @@ executed, but it never overrides the five canonical design files.
 **Generation workflow.** After a canonical edit:
 
 1. Run `./build/ConvertTo-HumanDocumentation.ps1` to regenerate detailed engine pages.
-2. Run `/agentkit:docs` to regenerate `docs/docs/guide.md` when the brief, design, or contract
-   changed semantically.
+2. Regenerate `docs/docs/guide.md` by hand when the brief, design, or contract changed
+   semantically — `/agentkit:docs` no longer exists.
 3. Run `./build/ConvertTo-HumanDocumentation.ps1 -StampGuide` after regenerating the guide.
 4. Run `./build/Test-Documentation.ps1`; it checks both exact generated pages and guide digest.
 
-Routing, when a command is run:
+**Routing, when a command is run:**
 
 | Command | Tier |
 |---|---|
-| `/agentkit:brief`, `/agentkit:design`, `/agentkit:spec`, `/agentkit:plan` | Opus, high |
+| `/agentkit:brief`, `/agentkit:design`, `/agentkit:plan`, `/agentkit:interview` | Opus, high |
 | `/agentkit:redteam` | strongest model, **different vendor from the design author**; if it must be Claude, a fresh Opus session |
-| `/agentkit:slice` | Sonnet, medium — high for a large or difficult work unit |
+| `/agentkit:next` | Sonnet, medium — high for a large or difficult work unit |
+| `/agentkit:fix` | Sonnet, medium — escalate to deep reasoning only to diagnose |
 | `/agentkit:align` | Opus, high to decide which side of a drift is correct; Sonnet, medium to apply the edits |
-| `/agentkit:install` | Sonnet, medium |
-| `/agentkit:install-all` | Sonnet, medium — escalate only to judge whether a per-repo hard stop is actually safe to resolve; never to resolve it unattended |
-| `/agentkit:docs` | Sonnet, medium — generate the developer guide from canonical `design/`, then stamp its digest |
-| `/agentkit:track` | Sonnet, medium — escalate only to judge whether a drifted work unit is a design change |
-| `/agentkit:check` | Sonnet, medium — escalate to deep reasoning only to diagnose a failure, never to run the gates |
-| `/agentkit:pr` | Sonnet, medium |
-| `/agentkit:resolve` | Sonnet, medium — escalate to judge a contested finding, not to triage the obvious ones |
-| `/agentkit:tune` | Sonnet, medium — never escalates; an architectural ask is routed to the command that owns it, not refined |
-| `/agentkit:help` | Haiku, low — orientation from file existence and a tracker listing; escalate only where the repository's state matches no stage |
-| `/agentkit:clean` | Haiku, low — mechanical git housekeeping; escalate only to judge whether an unmerged-looking branch is actually safe to delete |
-| `/agentkit:hold` | Sonnet, medium — `Frozen because`/`Lifts when` come from the user, never invented |
-| `/agentkit:resume` | Sonnet, medium for the sequencing; runs `/agentkit:align` (Opus, high) and `/agentkit:track` (Sonnet, medium) as its own phases |
-
-**`/agentkit:track` reads `design/30-slices.md`.** It opens one issue per unvisited W-numbered work unit;
-W is this repository's retained slice prefix. New units carry stable per-criterion ids
-(`W42.1`, `W42.2`, never reused or renumbered — the same discipline as the existing positional
-`W3a` unit numbering, one level down), and issues open in the shape the kit's `skills/track/SKILL.md`
-defines: a human-first narrative, `### Done when` checkboxes, and a fenced `<!-- agent:start -->`
-block that `/agentkit:track` regenerates but never overwrites outside the fence. Existing checked units
-predate this and are not retrofitted or reopened.
-
-**Tracking work.** Opening, labelling, closing, commenting on, and editing an issue — including
-one opened by someone else — needs no per-instance approval in a repository the user owns;
-issues are cheap and reversible. Creating a milestone or a project is carved out the same way;
-deleting either is not, since that direction is not cheaply reversible. Writing to a repository
-the user does not own is never carved out. **Resolving or replying to a review thread is not
-covered by this carve-out** — a pull request's review threads are a different object and stay
-authorized regardless, same as any other external write in **Git and Pull Requests** below.
-`/agentkit:track` owns every GitHub write it can make idempotent; closing an issue and ticking a
-checkbox are the exceptions — the command that observes the work done (`/agentkit:track`, `/agentkit:slice`)
-does those directly, in the same run, rather than waiting for a sync pass.
-
-**Session boundaries.** Routing above says which model runs a command. This says **when a session
-must end.** A boundary exists wherever carrying context would corrupt the next step's judgement, or
-wherever the next step must read the tree rather than remember it. The artifact is the handoff, not
-the conversation — a stage that writes one has already handed over everything the next stage is
-entitled to.
+| `/agentkit:install`, `/agentkit:install-all`, `/agentkit:install-review`, `/agentkit:sync` | Sonnet, medium — escalate only to judge whether a hard stop is safe to resolve |
+
+**Tracking work.** The kit no longer opens issues from `design/30-slices.md`; `design/state/work/`
+and `design/state-index.md` are an adopted-but-orphaned mirror, kept in place until the user
+decides whether to retire them. Opening, labelling, closing, commenting on, and editing an
+issue — including one opened by someone else — needs no per-instance approval in a repository
+the user owns; issues are cheap and reversible. Creating a milestone or a project is carved
+out the same way; deleting either is not. Writing to a repository the user does not own is
+never carved out. **Resolving or replying to a review thread is not covered by this
+carve-out** — review threads stay authorized regardless, same as any other external write in
+**Git and Pull Requests** below.
+
+**Session boundaries.** A boundary exists wherever carrying context would corrupt the next
+step's judgement. The artifact is the handoff, not the conversation.
 
 | Boundary | Rule | Why |
 |---|---|---|
-| `/agentkit:design` → `/agentkit:redteam` | **Fresh session, and a different vendor.** | A model recognises its own output distribution and defends it. Fresh context on the same model is already the weak form; the same session is not a review at all. |
-| Any stage that writes a canonical `design/` file → the next | Fresh. | The next stage's input is the committed file. A session that also remembers the arguments behind it will design against the arguments. |
-| `/agentkit:plan` → `/agentkit:slice` | Fresh, and **one work unit per session**. | A unit that does not fit one session without compaction is too large — that is a `/agentkit:plan` defect, so say so rather than pressing on. |
-| `/agentkit:slice` → `/agentkit:check` → `/agentkit:pr` → `/agentkit:resolve` | **Same session.** | These act on the branch and worktree the unit just produced, and `/agentkit:pr` must carry `/agentkit:check`'s did-not-run list into the description **verbatim**. A fresh session would restate it from a summary, which is the fabricated gate result **Validation** below exists to prevent. |
-| merge → `/agentkit:track` | Fresh. | `/agentkit:track` reads the tracker and `design/` as they now stand. The session that just implemented the unit holds an opinion about whether it is done, and doneness is the user's mark, not an agent's. |
-| implementation → `/agentkit:align` | Fresh. | It compares the tree against the canonical documents. The session that wrote the code carries what it *intended* to write, which is the one thing the comparison must not be given. |
-
-**Compaction is a boundary you did not choose.** If a session compacts mid-unit, report it — the
-unit was mis-sized, and the work after the compaction was done against a summary of the contract
-rather than the contract itself.
-
-**End a response that lands on a fresh-session boundary with a banner, not a footnote.** A
-boundary buried in the last sentence of a report gets carried into the next reply of the same
-session out of habit, which is the exact failure the boundary exists to prevent. Set it off
-visibly — a horizontal rule and a bold line is enough — naming: the boundary just crossed, the
-next command, and its tier from the routing table above. For example:
+| `/agentkit:design` → `/agentkit:redteam` | **Fresh session, and a different vendor.** | A model recognises its own output distribution and defends it. |
+| Any stage that writes a canonical `design/` file → the next | Fresh. | The next stage's input is the committed file, not the arguments behind it. |
+| `/agentkit:plan` → `/agentkit:next` | Fresh. | `/agentkit:next` reads the plan from `origin/<default branch>`, not from memory. |
+| `/agentkit:next` (within one run) | **Same session.** | It carries a slice from code to merged PR, and the PR's *Verified* section must state what did not run. |
+| implementation → `/agentkit:align` | Fresh. | It compares the tree against the canonical documents; the session that wrote the code carries what it *intended*. |
 
-```
----
-**Session boundary.** This context should not carry into `/agentkit:track`.
-Next: `/agentkit:track`, fresh session, Sonnet, medium.
----
-```
-
-Do not run the next command yourself. Ending a session may be the next step, and a command that
-starts work cannot also tell the user to start a new one for it — that restriction is unchanged,
-only how visibly the handoff is stated.
+**Compaction is a boundary you did not choose.** If a session compacts mid-slice, report it —
+the slice was mis-sized, and the work after it was done against a summary of the contract.
 
 ### Single Ownership
 
@@ -534,19 +481,14 @@ in CI, with nothing saying why.
 feature branches do not, so it is discipline rather than enforcement. If a pushed commit
 needs changing, add a follow-up commit.
 
-**Deleting a local branch `/agentkit:clean` independently confirms via `git branch --merged` is
-delegated in this repository.** `/agentkit:clean` (the kit's `skills/clean/SKILL.md`) runs proactively — as
-soon as a merge is on the table, not only when asked — and deletes every branch on that
-confirmed list without a chat confirmation first; the `--merged` check is the authorization.
-It also may stash (never discard) a dirty tree to unblock its own branch switch, and always
-reports the stash back rather than popping it silently. This delegation stops exactly where
-`--merged` stops: a branch it did not confirm, or a `-d` refusal on one it did, still needs a
-separate ask before anything stronger (`-D`) is even considered.
-
-**`/agentkit:slice` may push the branch it creates and open its PR as a draft, without asking** —
-carved out of the authorization rule the same as an issue: a draft blocks no one and requests
-no review, so opening one carries the same reversibility argument. Marking that PR ready for
-review, and merging it, are not carved out and stay `/agentkit:pr`'s and the user's respectively.
+**Housekeeping is delegated.** `/agentkit:next` runs the kit's `invoke-housekeeping.ts`, which deletes a
+local branch only after confirming it merged (`git branch --merged`, or a squash-merged PR whose
+head matches the local tip), never stashes, and escalates anything else. That check is the
+authorization; a branch it did not confirm needs a separate ask before `-D` is considered.
+
+**`/agentkit:next` opens its pull requests non-draft and merges them** through the kit's merge script,
+which refuses on unconfirmed checks. The earlier draft-PR carve-out belonged to `/agentkit:slice`,
+which no longer exists.
 
 **Do not enable auto-merge.** Open the PR, report the check outcomes, and leave the merge to
 the user. (Auto-merge is enabled at the repository level and `required_approving_review_count`
diff --git a/design/90-decisions.md b/design/90-decisions.md
index b0f9c8c4..88911103 100644
--- a/design/90-decisions.md
+++ b/design/90-decisions.md
@@ -2211,3 +2211,9 @@ Reversibility: cheap. Restoring the copies is one checkout of `6a91aec`'s `.clau
 Context: `.claude/settings.json` ran `pwsh … tools/Measure-Session.ps1` on `SessionEnd`, but that script no longer exists — it left this repository when the kit moved to a single home install, and the kit later ported it to Node as `measure-session.ts` — so the hook failed at the end of every session. The kit's setup installs one global `SessionEnd` hook in `~/.claude/settings.json` that logs every project.
 Chosen: remove this repository's `hooks.SessionEnd` entry, in the AgentKit sync to `v2026.10.06.1`. Nothing else in `settings.json` changes.
 Rejected: point it at `measure-session.ts` — the global hook already runs that script, so every session would be logged twice; leave it — it keeps failing at every session end.
+
+### 2026-10-07 — Kit sync to `v2026.10.07`: the track/slice pipeline is retired, and the docs stop naming it
+Context: The kit's `v2026.10.07` ships thirteen commands (`interview`, `design`, `plan`, `next`, `fix`, `align`, `redteam`, `install`, `install-all`, `install-review`, `sync`, `brief`, `agentkit`) and no longer has `slice`, `check`, `pr`, `resolve`, `track`, `docs`, `hold`, `resume`, `help`, `tune`, `spec` or `clean`, nor `Test-DesignState.ps1` or `Update-DesignProjection.ps1`. `CLAUDE.md` and `AGENTS.md` still documented that pipeline, the `skills/<name>/SKILL-local.md` companions and the `design/state/` projection as live.
+Chosen: Delete the 14 `SKILL-local.md` files and the three kit-owned `tools/*.ps1` copies (`Test-DesignDrift.ps1` and its test, `Test-CIWorkflow.Tests.ps1`). Rewrite `CLAUDE.md`'s Design Freeze and Agent Kit sections and `AGENTS.md`'s design-state section to the current command set. The freeze stays, as a manual convention: `design/FROZEN.md` is written and lifted by hand. `design/state/` and `design/state-index.md` stay in place, orphaned, until a separate decision retires them.
+Rejected: **Leave the docs as written** — they point readers at commands that do not exist. **Retire the freeze with the commands that enforced it** — the reasoning for it (no fixed point while implementation is the bottleneck) still holds and a marker costs nothing. **Delete `design/state/` now** — that is a decision about the tracker mirror, not a transcription, and `design/20-contract.md`'s Design-state mechanics section names it. **Edit `design/20-contract.md` here** — it carries the same stale references (`Test-DesignState.ps1`, `Update-DesignProjection.ps1`, `/track`) and belongs to a `/agentkit:align` pass, not an install.
+Reversibility: cheap. The deletions and the prose are one revert of this change.
diff --git a/skills/align/SKILL-local.md b/skills/align/SKILL-local.md
deleted file mode 100644
index f9724983..00000000
--- a/skills/align/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here.
diff --git a/skills/check/SKILL-local.md b/skills/check/SKILL-local.md
deleted file mode 100644
index 7ba68568..00000000
--- a/skills/check/SKILL-local.md
+++ /dev/null
@@ -1,30 +0,0 @@
-## gate-commands
-| Flagged step | Workflow | Run locally |
-|---|---|---|
-| `Typecheck` | `ci.yml` | `npm --prefix src/engine run typecheck` |
-| `Lint` | `ci.yml` | `npm --prefix src/engine run lint` |
-| `Test` | `ci.yml` | `npm --prefix src/engine test` |
-| `Pack package` | `ci.yml` | `cd src/engine; npm pack --silent` |
-| `Inspect tarball` | `ci.yml` | `tar -tzf` the packed tarball; assert no `src/`, no `tsconfig*.json`, no `.test.*` artifacts, `package/dist/` present |
-| `Consumer smoke` | `ci.yml` | Clear `consumer-smoke/{node_modules,package-lock.json,dist}`, then `npm run install:engine && npm run build && npm run smoke` |
-| `Parse-check PowerShell scripts` | `verify.yml` | Parse every `*.ps1` with `[System.Management.Automation.Language.Parser]::ParseFile`, as the step does |
-| `Run Pester tests` | `verify.yml` | `Invoke-Pester -Path tools -Output Detailed -PassThru` |
-| `Validate Markdown links, terminology, and generated files` | `docs-ci.yml` | `./build/Test-Documentation.ps1` |
-| `Build documentation` | `docs-ci.yml` | `./docs.ps1 -BuildOnly` — needs Docker **and** an installed `docs.ps1` |
-| `Build and verify landing page` | `docs-ci.yml` | `npm --prefix src/engine run build; npm --prefix site run check` |
-| `Merge landing page into documentation build` | `docs-ci.yml` | `npm --prefix site run merge` — needs a completed docs build in `artifacts/docs` |
-| `Test the host` | `host-image.yml` | `dotnet test src/host/SubZeroDev.GameEngine.Host.Tests/…csproj` — needs `NUGET_GITHUB_TOKEN` for the sibling-repo feed |
-| `Positive route and probe smoke` | `host-image.yml` | Build and run the host image, then `curl` `/`, `/roadmap/`, `/docs/`, `/health/live`, `/health/ready` → 200 and an unknown route → 404 |
-| `Negative fixture -- corrupted artifact must fail to start` | `host-image.yml` | `docker build -f tools/host-smoke/Dockerfile.negative-fixture …`; the run must exit non-zero |
-
-Discover by reading the workflow files rather than trusting a memorized list; the fifteen rows
-above describe this repository's gates as of the last sync (2026-08-21), not a substitute for
-reading `.github/workflows/*.yml` directly. `Test-Companion.ps1` and `Test-DesignState.ps1` are
-no longer `# verification: true` flagged in `verify.yml` — worth running, but not required gates.
-
-Local discovery, beyond the workflow files:
-```powershell
-if (Test-Path package.json) { Get-Content package.json | Select-String '"scripts"' -Context 0,20 }
-Get-ChildItem . -Include *.sln, *.csproj -Recurse -Depth 2
-if (Test-Path build) { Get-ChildItem build -Filter *.ps1 }
-```
diff --git a/skills/design/SKILL-local.md b/skills/design/SKILL-local.md
deleted file mode 100644
index f9724983..00000000
--- a/skills/design/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here.
diff --git a/skills/help/SKILL-local.md b/skills/help/SKILL-local.md
deleted file mode 100644
index 1c12041f..00000000
--- a/skills/help/SKILL-local.md
+++ /dev/null
@@ -1,8 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here. Slice ids use the retained `W<n>` scheme, not `S<n>` —
-read `/slice S3` below as `/slice W<n>`.
-
-## document-map
-`design/` here is a compound canonical set with generated `human-doc` blocks; file existence
-alone is a weaker stage signal than in a bare repository. Read the marked delivery ledger in
-`design/30-slices.md` rather than inferring status from the generated `docs/docs/engine/TODO.md`.
diff --git a/skills/hold/SKILL-local.md b/skills/hold/SKILL-local.md
deleted file mode 100644
index 909061a6..00000000
--- a/skills/hold/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here; the section is *The Design Freeze*.
diff --git a/skills/plan/SKILL-local.md b/skills/plan/SKILL-local.md
deleted file mode 100644
index d4976240..00000000
--- a/skills/plan/SKILL-local.md
+++ /dev/null
@@ -1,5 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here. This repository retains its established `W` prefix — do
-not renumber existing units to `S`; every `S<n>` heading and id below (`S<n>.1`, `S3.1`, …) reads
-as `W<n>` here. Existing per-unit numbering already goes one level deeper than a bare `.1`/`.2`
-(e.g. `W3a`) — extend that scheme rather than introducing a second one.
diff --git a/skills/pr/SKILL-local.md b/skills/pr/SKILL-local.md
deleted file mode 100644
index 0384e1b6..00000000
--- a/skills/pr/SKILL-local.md
+++ /dev/null
@@ -1,10 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here. Slice ids use the retained `W<n>` scheme, not `S<n>` — read
-the `Slice: S3` example below as `Unit: W<n>`.
-
-## tightened-authorization
-`CLAUDE.md`, *Git and Pull Requests* carves `/slice` out to open its PR as a draft, without asking
-— narrower than this core's "never as a draft." Where `gh pr view` finds a PR already open on this
-branch, **ask before marking it ready for review** rather than treating the existing-PR case as
-already actionable; do not open a second PR either way. This tightens what the core does on its
-own — it does not remove the core's authorization to open a PR at all when none exists.
diff --git a/skills/redteam/SKILL-local.md b/skills/redteam/SKILL-local.md
deleted file mode 100644
index f9724983..00000000
--- a/skills/redteam/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here.
diff --git a/skills/resolve/SKILL-local.md b/skills/resolve/SKILL-local.md
deleted file mode 100644
index 00ee07da..00000000
--- a/skills/resolve/SKILL-local.md
+++ /dev/null
@@ -1,3 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here; the section this file cites as *Git and delivery* is
-*Git and Pull Requests*, and *Working with me* is *Working Conventions*.
diff --git a/skills/resume/SKILL-local.md b/skills/resume/SKILL-local.md
deleted file mode 100644
index f9724983..00000000
--- a/skills/resume/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here.
diff --git a/skills/slice/SKILL-local.md b/skills/slice/SKILL-local.md
deleted file mode 100644
index 771c281e..00000000
--- a/skills/slice/SKILL-local.md
+++ /dev/null
@@ -1,8 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here. Slice ids use the retained `W<n>` scheme, not `S<n>` —
-`slice/S<n>`, `S3.1`, and the issue-title match below all read with `W<n>` in place of `S<n>`.
-
-## document-map
-`design/` here is a compound canonical set with generated `human-doc` blocks; read the marked
-delivery ledger in `design/30-slices.md` in full, not the generated `docs/docs/engine/TODO.md`
-copy.
diff --git a/skills/spec/SKILL-local.md b/skills/spec/SKILL-local.md
deleted file mode 100644
index f9724983..00000000
--- a/skills/spec/SKILL-local.md
+++ /dev/null
@@ -1,2 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here.
diff --git a/skills/track/SKILL-local.md b/skills/track/SKILL-local.md
deleted file mode 100644
index 35a0bd30..00000000
--- a/skills/track/SKILL-local.md
+++ /dev/null
@@ -1,24 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here. This repository retains its established `W` prefix — every
-`S<n>` heading, id (`S3.1`), and issue-title pattern (`S<n> —`) below reads as `W<n>` here, matching
-the existing positional `W3a` unit numbering one level down.
-
-## document-map
-`design/30-slices.md` has no `## Outstanding` / `## Landed` split, and is not going to acquire
-one. A unit's state is carried on its own heading — `### [x] W102 — …` for landed, `### [ ] W103
-— …` for outstanding, `### [~] W68 — … — cancelled` for abandoned — inside the topical section
-that explains why the unit exists. `Test-DesignDrift.ps1` reads that convention correctly; it is
-the *retirement* step that does not, because there is no second section for a body to move into.
-
-Read this document by heading marker, therefore, and treat *Landed slices → retired* as not
-applying here: `tools/Update-SlicesDocument.ps1` exits 2 with `NoLandedSection` on every run, and
-that is the expected outcome rather than a finding to report or a gap to repair. Nothing else
-about the step changes — a landed unit is still one whose issue is closed, and the reported-not-
-claimed rule for a step that did not run is unaffected.
-
-**Why the document is shaped this way.** The `[x]` heading and the surrounding narrative are the
-same artifact: `### Correctness Debt — The Tick Pipeline Runs Twenty Systems and Implements
-Fifteen` is only legible with the five units under it, landed or not, in delivery order.
-Retiring a body to a table elsewhere would leave the section that argued for it empty, and
-restructuring the ~4,600 lines would rewrite every `{#w<n>}` anchor and every cross-link into
-them across `design/` and the generated pages. Recorded in `design/90-decisions.md`, 2026-09-05.
diff --git a/skills/tune/SKILL-local.md b/skills/tune/SKILL-local.md
deleted file mode 100644
index e56185ca..00000000
--- a/skills/tune/SKILL-local.md
+++ /dev/null
@@ -1,6 +0,0 @@
-## vocabulary
-`AGENTS.md` is named `CLAUDE.md` here — every `AGENTS.md §` citation this command emits should
-resolve through that file's real section names (e.g. the routing table lives under *The Agent
-Kit — Canonical Workflow*, not a section literally named "Command routing"); never emit a section
-name you have not seen in `CLAUDE.md` itself. Slice ids use the retained `W<n>` scheme, so
-`/slice S<n>` in the routing table reads `/slice W<n>` here.
diff --git a/tools/Test-CIWorkflow.Tests.ps1 b/tools/Test-CIWorkflow.Tests.ps1
deleted file mode 100644
index 0f09e716..00000000
--- a/tools/Test-CIWorkflow.Tests.ps1
+++ /dev/null
@@ -1,54 +0,0 @@
-#Requires -Version 7.0
-#Requires -Modules Pester
-
-<#
-  Regression coverage for #79: the "Run Pester tests" CI step calls the real design-state
-  check (S12.5, tools/Test-DesignState.Tests.ps1) against this repository, which needs an
-  authenticated gh exactly as the later "Check the design state against the tree" step
-  already does - so it needs the same GH_TOKEN. Without it, an unauthenticated gh turns
-  S12.5 into a could-not-evaluate (TrackerUnavailable) rather than a check of anything this
-  step is meant to gate.
-
-  This regression only applies once the kit's design-state tracking is adopted here and
-  verify.yml has grown the "Check the design state against the tree" step this test compares
-  against - the kit's own compatibility promise (design/90-decisions.md, 2026-08-19) does not
-  migrate that step to installed targets, so its absence here is not a divergence to report.
-#>
-
-Describe 'CI workflow: the Run Pester tests step is authenticated (#79)' {
-    BeforeAll {
-        # The assertion itself is intentionally skipped until this repository
-        # adopts the design-state workflow step.  Pester still evaluates
-        # BeforeAll for skipped examples, so avoid reading a workflow that is
-        # deliberately absent in that compatibility state.
-        $workflowPath = Join-Path (Split-Path $PSScriptRoot -Parent) '.github/workflows/verify.yml'
-        $script:RunGhTokenTest = (Test-Path $workflowPath) -and
-            (Select-String -LiteralPath $workflowPath -Pattern '- name: Check the design state against the tree' -Quiet)
-        if (-not $script:RunGhTokenTest) {
-            $script:Lines = @()
-            return
-        }
-
-        $script:WorkflowPath = $workflowPath
-        $script:Lines = Get-Content -LiteralPath $script:WorkflowPath
-    }
-
-    It 'the "Run Pester tests" step carries a GH_TOKEN env, the same as "Check the design state against the tree"' {
-        if (-not $script:RunGhTokenTest) {
-            Set-ItResult -Skipped -Because 'This repository has not adopted the design-state workflow step.'
-            return
-        }
-        $stepIndex = ($script:Lines | Select-String -Pattern '- name: Run Pester tests').LineNumber
-        $stepIndex | Should -Not -BeNullOrEmpty
-
-        # The step body runs from its `- name:` line to the line before the next `- name:`
-        # (or end of file), so this only inspects this one step's own env block.
-        $nextStepIndex = ($script:Lines | Select-String -Pattern '^\s*- name:' |
-            Where-Object { $_.LineNumber -gt $stepIndex } |
-            Select-Object -First 1).LineNumber
-        $endIndex = if ($nextStepIndex) { $nextStepIndex - 1 } else { $script:Lines.Count }
-        $stepBody = $script:Lines[($stepIndex - 1)..($endIndex - 1)] -join "`n"
-
-        $stepBody | Should -Match 'GH_TOKEN:\s*\$\{\{\s*secrets\.GITHUB_TOKEN\s*\}\}'
-    }
-}
diff --git a/tools/Test-DesignDrift.Tests.ps1 b/tools/Test-DesignDrift.Tests.ps1
deleted file mode 100644
index 804af536..00000000
--- a/tools/Test-DesignDrift.Tests.ps1
+++ /dev/null
@@ -1,347 +0,0 @@
-#Requires -Version 7.0
-#Requires -Modules Pester
-
-<#
-  Test-DesignDrift.ps1 exits the process on every path (0/1/2), which would kill the Pester
-  runner if invoked with `&`. Same structure as Wait-PullRequestCheck.Tests.ps1: the script
-  guards its exit-calling wrapper with `$MyInvocation.InvocationName -ne '.'`, so dot-sourcing
-  defines its functions here and skips the wrapper. `Invoke-DriftCheck` is called directly and
-  asserted on its returned result; `Get-DriftExitCode` is a pure state->code map tested alone.
-
-  The two boundaries - the tracker and git - are mocked at their own functions rather than at
-  `gh` and `git`, because both read a native *exit code* as their answer and a mock cannot set
-  $LASTEXITCODE. Mocking the seam above them tests the comparison, which is what this script is.
-
-  Every slices document is written into $TestDrive; none of these tests read the real one, so
-  they do not start failing when a slice lands.
-#>
-
-BeforeAll {
-    $script:ScriptPath = Join-Path $PSScriptRoot 'Test-DesignDrift.ps1'
-    # The dot-source runs the script's own Set-StrictMode/$ErrorActionPreference in this scope.
-    # Captured here and restored in the matching AfterAll so they do not leak into whichever
-    # test file Pester runs next in the same invocation.
-    $script:PreDotSourceErrorActionPreference = $ErrorActionPreference
-    . $script:ScriptPath
-
-    function New-SlicesDoc {
-        param([Parameter(Mandatory)][string] $Content, [string] $Name = 'slices.md')
-        $path = Join-Path $TestDrive $Name
-        Set-Content -LiteralPath $path -Value $Content -Encoding utf8
-        $path
-    }
-
-    function New-Issue {
-        param([int] $Number, [string] $Title, [string] $Body)
-        [pscustomobject]@{ number = $Number; title = $Title; state = 'OPEN'; body = $Body }
-    }
-
-    function New-Tracker {
-        param([object[]] $Issues = @())
-        [pscustomobject]@{ Issues = @($Issues); Failure = $null }
-    }
-
-    # One slice, two criteria - the shape every positive case starts from. This repository's
-    # own vocabulary (W, not S; a `### [ ]` heading, not `##`) per track-local.md.
-    $script:TwoCriterionDoc = @'
-# Slices
-
-## Core
-
-### [ ] W1 — A slice
-
-Delivers: something a reader can follow.
-
-- **Done when:**
-  - W1.1 The first criterion holds.
-  - W1.2 The second criterion holds.
-'@
-}
-
-AfterAll {
-    $ErrorActionPreference = $script:PreDotSourceErrorActionPreference
-    Set-StrictMode -Off
-}
-
-Describe 'Test-DesignDrift' {
-
-    Context 'criterion ids' {
-
-        It 'matching ids on both sides is Clean, exit 0' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body "### Done when`n- [ ] **W1.1** first`n- [x] **W1.2** second"
-            ) }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'Clean'
-            $r.Findings.Count | Should -Be 0
-            $r.SlicesCompared | Should -Be 1
-            Get-DriftExitCode -State $r.State | Should -Be 0
-        }
-
-        It 'reworded criteria with the same ids are not drift' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body "- [ ] **W1.1** completely different wording here`n- [ ] **W1.2** and here too"
-            ) }
-
-            (Invoke-DriftCheck -SlicesPath $path).State | Should -Be 'Clean'
-        }
-
-        It 'a criterion the doc has already ticked is still a criterion, not a removal' {
-            # This repository ticks criteria in place, so the doc-side id sits behind a
-            # `- [x] ` the issue-side regex has always tolerated and this one had not. Left
-            # unhandled it reads as InIssueNotDoc - a renumber - which is the one finding
-            # that can invalidate a tick.
-            $path = New-SlicesDoc -Content @'
-# Slices
-
-## Core
-
-### [ ] W1 — A slice
-
-Delivers: something a reader can follow.
-
-- **Done when:**
-  - [x] W1.1 The first criterion holds, and has landed.
-  - W1.2 The second criterion holds.
-'@
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body "- [x] **W1.1** first`n- [ ] **W1.2** second"
-            ) }
-
-            (Invoke-DriftCheck -SlicesPath $path).State | Should -Be 'Clean'
-        }
-
-        It 'an id in the doc but not the issue is reported as InDocNotIssue, exit 1' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body '- [ ] **W1.1** first'
-            ) }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'Drifted'
-            $r.Findings.Kind | Should -Contain 'InDocNotIssue'
-            ($r.Findings | Where-Object Kind -eq 'InDocNotIssue').Detail | Should -Be 'W1.2'
-            Get-DriftExitCode -State $r.State | Should -Be 1
-        }
-
-        It 'an id in the issue but not the doc - a renumber - is reported as InIssueNotDoc' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body "- [ ] **W1.1** first`n- [x] **W1.7** a tick now pointing at something else"
-            ) }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'Drifted'
-            ($r.Findings | Where-Object Kind -eq 'InIssueNotDoc').Detail | Should -Be 'W1.7'
-        }
-
-        It 'an id cited in prose outside a slice section is not counted as a criterion' {
-            # The real document does exactly this in its Contract questions section, so a
-            # whole-file regex would invent criteria that were never cut.
-            $path = New-SlicesDoc -Content @'
-# Slices
-
-## Contract questions
-
-None outstanding - zero checks yields NotEvaluated, exercised by W1.9, and the batch by W1.8.
-
-### [ ] W1 — A slice
-
-- **Done when:**
-  - W1.1 The only real criterion.
-'@
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body '- [ ] **W1.1** the only real criterion'
-            ) }
-
-            (Invoke-DriftCheck -SlicesPath $path).State | Should -Be 'Clean'
-        }
-
-        It 'a landed slice with no body is not reported as a removal' {
-            # The kit-generic `## Landed` table layout. This repository does not retire
-            # slice bodies this way (every `### [x] W<n>` keeps its full "Done when" list),
-            # but the table-row path stays supported and is worth covering.
-            $path = New-SlicesDoc -Content @'
-# Slices
-
-## Outstanding
-
-None.
-
-## Landed
-
-| Slice | Name | Issue | Criteria | Body complete at |
-|---|---|---|---|---|
-| **W1** | A slice that landed | #9, closed | W1.1-W1.2 | `af610a6` |
-'@
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice that landed' -Body "- [x] **W1.1** first`n- [x] **W1.2** second"
-            ) }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'Clean'
-            $r.Findings.Count | Should -Be 0
-            $r.SlicesCompared | Should -Be 0
-        }
-
-        It 'a slice with no issue at all is reported rather than skipped' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue { New-Tracker }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'Drifted'
-            $r.Findings.Kind | Should -Contain 'NoIssue'
-        }
-
-        It 'a ticked slice with no issue is not reported, while an unticked one still is' {
-            # This repository marks a slice landed with `### [x]` and keeps its body, rather
-            # than retiring it into the `## Landed` table. track.md does not sync a landed
-            # slice, so there is no issue to open for one that predates the tracker - but the
-            # suppression must not swallow an outstanding slice that genuinely lacks one.
-            $path = New-SlicesDoc -Content @'
-# Slices
-
-## Core
-
-### [x] W1 — A slice that landed before the tracker
-
-- **Done when:**
-  - [x] W1.1 The first criterion holds.
-
-### [ ] W2 — A slice still outstanding
-
-- **Done when:**
-  - W2.1 The first criterion holds.
-'@
-            Mock Get-TrackerIssue { New-Tracker }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            @($r.Findings | Where-Object { $_.Kind -eq 'NoIssue' }).Slice | Should -Be @('W2')
-        }
-    }
-
-    Context 'pin ancestry' {
-
-        BeforeEach {
-            $script:PinnedDoc = New-SlicesDoc -Content $script:TwoCriterionDoc -Name 'pinned.md'
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' `
-                    -Body "- [ ] **W1.1** first`n- [ ] **W1.2** second`n<!-- agent:start -->`nScope and criteria: ``design/30-slices.md`` § W1 @ ``deadbee```n<!-- agent:end -->"
-            ) }
-        }
-
-        It 'a pin that is an ancestor of HEAD is Clean' {
-            Mock Test-CommitIsAncestor { 'Ancestor' }
-
-            $r = Invoke-DriftCheck -SlicesPath $script:PinnedDoc
-
-            $r.State | Should -Be 'Clean'
-            Should -Invoke Test-CommitIsAncestor -Exactly -Times 1
-        }
-
-        It 'a pin that is not an ancestor is drift, exit 1' {
-            Mock Test-CommitIsAncestor { 'NotAncestor' }
-
-            $r = Invoke-DriftCheck -SlicesPath $script:PinnedDoc
-
-            $r.State | Should -Be 'Drifted'
-            ($r.Findings | Where-Object Kind -eq 'PinNotAncestor').Detail | Should -Be 'deadbee'
-            Get-DriftExitCode -State $r.State | Should -Be 1
-        }
-
-        It 'a pin this clone cannot resolve is NOT clean and NOT drift - it is exit 2' {
-            # The distinction I12 exists for: absence of an answer never becomes an answer.
-            Mock Test-CommitIsAncestor { 'Unresolvable' }
-
-            $r = Invoke-DriftCheck -SlicesPath $script:PinnedDoc
-
-            $r.State | Should -Be 'NotEvaluated'
-            $r.Findings.Count | Should -Be 0
-            $r.Failures.Reason | Should -Contain 'PinUnresolvable'
-            Get-DriftExitCode -State $r.State | Should -Be 2
-        }
-    }
-
-    Context 'incomplete runs never report clean' {
-
-        It 'an unreadable tracker is NotEvaluated, exit 2' {
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc
-            Mock Get-TrackerIssue {
-                [pscustomobject]@{ Issues = @(); Failure = (New-Failure -Reason 'GhUnavailable' -Detail 'gh exited 4') }
-            }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'NotEvaluated'
-            $r.Failures.Reason | Should -Contain 'GhUnavailable'
-            Get-DriftExitCode -State $r.State | Should -Be 2
-        }
-
-        It 'a missing slices document is NotEvaluated, and the tracker is never read' {
-            Mock Get-TrackerIssue { New-Tracker }
-
-            $r = Invoke-DriftCheck -SlicesPath (Join-Path $TestDrive 'nothing-here.md')
-
-            $r.State | Should -Be 'NotEvaluated'
-            $r.Failures.Reason | Should -Contain 'SlicesDocMissing'
-            Should -Invoke Get-TrackerIssue -Exactly -Times 0
-        }
-
-        It 'a criterion numbered for another slice is unparseable, not silently filed' {
-            $path = New-SlicesDoc -Content @'
-# Slices
-
-### [ ] W1 — A slice
-
-- **Done when:**
-  - W1.1 The first criterion.
-  - W2.4 Numbered for a slice this is not.
-'@ -Name 'stray.md'
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' -Body '- [ ] **W1.1** first'
-            ) }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.State | Should -Be 'NotEvaluated'
-            $r.Failures.Reason | Should -Contain 'UnparseableCriterion'
-        }
-
-        It 'drift AND a failed comparison resolves to exit 2, never 1' {
-            # 2 takes precedence: a run that found drift and also failed to finish is an
-            # incomplete run, and reporting it as a finished one is the failure I12 forbids.
-            $path = New-SlicesDoc -Content $script:TwoCriterionDoc -Name 'both.md'
-            Mock Get-TrackerIssue { New-Tracker -Issues @(
-                New-Issue -Number 9 -Title 'W1 — A slice' `
-                    -Body "- [ ] **W1.1** first`n``design/30-slices.md`` § W1 @ ``deadbee``"
-            ) }
-            Mock Test-CommitIsAncestor { 'Unresolvable' }
-
-            $r = Invoke-DriftCheck -SlicesPath $path
-
-            $r.Findings.Kind | Should -Contain 'InDocNotIssue'
-            $r.Failures.Reason | Should -Contain 'PinUnresolvable'
-            $r.State | Should -Be 'NotEvaluated'
-            Get-DriftExitCode -State $r.State | Should -Be 2
-        }
-    }
-
-    Context 'exit code map' {
-
-        It 'maps each state, and refuses an unknown one rather than defaulting to 0' {
-            Get-DriftExitCode -State 'Clean'        | Should -Be 0
-            Get-DriftExitCode -State 'Drifted'      | Should -Be 1
-            Get-DriftExitCode -State 'NotEvaluated' | Should -Be 2
-            { Get-DriftExitCode -State 'Something' } | Should -Throw
-        }
-    }
-}
diff --git a/tools/Test-DesignDrift.ps1 b/tools/Test-DesignDrift.ps1
deleted file mode 100644
index aaf5817c..00000000
--- a/tools/Test-DesignDrift.ps1
+++ /dev/null
@@ -1,359 +0,0 @@
-#Requires -Version 7.0
-<#
-.SYNOPSIS
-    Reports drift between design/30-slices.md and the GitHub tracker, and changes neither.
-
-.DESCRIPTION
-    Two comparisons that a model should never do by eye, because both are set arithmetic over
-    files and both fail silently when done from memory (AGENTS.md, "What should stop being
-    model work" - the red row):
-
-      1. Criterion ids. Every `W<n>.<m>` under a slice's `Done when:` lines, against every
-         `W<n>.<m>` checkbox in that slice's issue. Reworded criteria are not drift; an added,
-         removed, or renumbered id is. A renumber is the finding that matters, because an
-         existing ticked checkbox then refers to something else.
-
-      2. Pin ancestry. Every `design/30-slices.md § W<n> @ <sha>` pin in an issue's agent
-         block, against `git merge-base --is-ancestor <sha> HEAD`. This is the check
-         design/90-decisions.md (2026-08-10) records as having been claimed done and not
-         been: fourteen of seventeen issues cited a commit that was not an ancestor of main,
-         and nothing in the repository could tell anyone that.
-
-    Read-only on both sides, by contract (I13). Which side of a drift is wrong is the user's
-    call, so this establishes only that the two disagree.
-
-    Exit codes: 0 no drift, 1 drift found, 2 could not evaluate. **2 takes precedence over 1**
-    - a run that found drift *and* failed to complete a comparison is an incomplete run, and
-    reporting it as a finished one is exactly the failure I12 forbids. Never prompts.
-
-.PARAMETER SlicesPath
-    Path to the slices document. Defaults to design/30-slices.md beside this script's repo root.
-
-.PARAMETER Repository
-    owner/repo. Defaults to the current git remote, via gh's own resolution.
-
-.PARAMETER Quiet
-    Suppresses the human-readable report only. The result object is always emitted.
-
-.EXAMPLE
-    pwsh ./tools/Test-DesignDrift.ps1
-#>
-[CmdletBinding()]
-param(
-    [string] $SlicesPath,
-    [string] $Repository,
-    [switch] $Quiet
-)
-
-Set-StrictMode -Version Latest
-$ErrorActionPreference = 'Stop'
-
-# Both comparisons below read a native command's *exit code* as their answer:
-# `git merge-base --is-ancestor` returns 1 to mean "not an ancestor", which is a finding, not
-# an error. On PowerShell 7.3+ this preference defaults to $true, which would turn that 1 into
-# a terminating error under $ErrorActionPreference = 'Stop' and report every stale pin as
-# unresolvable instead. Assigning it is inert on versions that predate it.
-$PSNativeCommandUseErrorActionPreference = $false
-
-function New-DriftResult {
-    param(
-        [string]   $State,
-        [object[]] $Findings = @(),
-        [object[]] $Failures = @(),
-        [int]      $SlicesCompared = 0
-    )
-    [pscustomobject]@{
-        State          = $State
-        Findings       = @($Findings)
-        Failures       = @($Failures)
-        SlicesCompared = $SlicesCompared
-    }
-}
-
-function New-Finding {
-    param([string]$Kind, [string]$Slice, [string]$Detail, [int]$Issue)
-    [pscustomobject]@{ Kind = $Kind; Slice = $Slice; Detail = $Detail; Issue = $Issue }
-}
-
-function New-Failure {
-    param([string]$Reason, [string]$Detail)
-    [pscustomobject]@{ Reason = $Reason; Detail = $Detail }
-}
-
-<#
-    Criterion ids are read only from lines under a `### [ ] W<n>` heading, never from the
-    whole file. Prose cites ids too - "exercised by W1.10" would appear in this document's
-    own cross-references - and a whole-file regex would count those as criteria that exist.
-    This repository retains its established `W` slice prefix in place of the kit's generic
-    `S` (design/30-slices.md, "How this document is kept"; skills/track/SKILL-local.md).
-#>
-function Get-SliceCriteria {
-    param([Parameter(Mandatory)][string] $Path)
-
-    if (-not (Test-Path -LiteralPath $Path)) {
-        return [pscustomobject]@{
-            Slices  = @{}
-            Landed  = @()
-            Failure = (New-Failure -Reason 'SlicesDocMissing' -Detail $Path)
-        }
-    }
-
-    $slices  = @{}
-    $landed  = [System.Collections.Generic.List[int]]::new()
-    $current = $null
-
-    foreach ($line in (Get-Content -LiteralPath $Path)) {
-        if ($line -match '^##\s' -and $line -notmatch '^###') {
-            # A thematic `##` section boundary (this repository groups slices under
-            # `## Core`, `## Post-MVP — Depth`, etc., not a flat `## Outstanding`) always
-            # ends the previous slice's body, the same as a new slice heading does below.
-            $current = $null
-            continue
-        }
-
-        if ($line -match '^###\s+\[') {
-            # A new third-level `### [ ] W<n>` heading always ends the previous slice's body,
-            # so a Done-when line can never be attributed across a section boundary.
-            $current = if ($line -match '^###\s+\[(?<mark>.)\]\s+W(?<n>\d+)\b') { [int]$Matches['n'] } else { $null }
-            if ($null -ne $current) {
-                # A ticked heading is how this repository marks a slice landed - it keeps the
-                # body rather than retiring it into the kit's `## Landed` table below, so the
-                # table path alone left $landed permanently empty here and every finished
-                # slice was compared as though it were still outstanding.
-                if ($Matches['mark'] -in @('x', 'X')) { $landed.Add($current) }
-                if (-not $slices.ContainsKey($current)) {
-                    $slices[$current] = [System.Collections.Generic.List[string]]::new()
-                }
-            }
-            continue
-        }
-
-        if ($line -match '^\|\s*\*\*W(?<n>\d+)\*\*\s*\|') {
-            $landed.Add([int]$Matches['n'])
-            continue
-        }
-
-        # The optional `[ ]`/`[x]` is not decoration either: this repository ticks individual
-        # criteria in place (`- [x] W103.2 ...`), so a regex requiring the id straight after
-        # the dash reads a completed criterion as missing from the doc and reports it as
-        # InIssueNotDoc - three false findings on W103 before this allowance existed.
-        if ($null -ne $current -and $line -match '^\s*-\s+(?:\[[ xX]\]\s+)?W(?<n>\d+)\.(?<m>\d+)\b') {
-            if ([int]$Matches['n'] -ne $current) {
-                # An id numbered for a different slice than the section it sits in. Reported
-                # rather than silently filed under either, because it is a defect in the doc.
-                if (-not $slices.ContainsKey(-1)) {
-                    $slices[-1] = [System.Collections.Generic.List[string]]::new()
-                }
-                $slices[-1].Add("W$($Matches['n']).$($Matches['m']) (found under W$current)")
-                continue
-            }
-            $slices[$current].Add("W$($Matches['n']).$($Matches['m'])")
-        }
-    }
-
-    [pscustomobject]@{
-        Slices  = $slices
-        Landed  = @($landed)
-        Failure = $null
-    }
-}
-
-function Get-IssueCriteria {
-    param([string] $Body)
-    if ([string]::IsNullOrWhiteSpace($Body)) { return @() }
-
-    $ids = [System.Collections.Generic.List[string]]::new()
-    foreach ($line in ($Body -split "`r?`n")) {
-        if ($line -match '^\s*-\s*\[[ xX]\]\s*\*{0,2}W(?<n>\d+)\.(?<m>\d+)\*{0,2}') {
-            $ids.Add("W$($Matches['n']).$($Matches['m'])")
-        }
-    }
-    @($ids)
-}
-
-function Get-IssuePin {
-    param([string] $Body)
-    if ([string]::IsNullOrWhiteSpace($Body)) { return $null }
-    # The backtick after `.md` is not optional decoration: track.md's pin format is
-    # `design/30-slices.md` § W3 @ `a1b2c3d`, so the path is code-fenced and the closing
-    # fence sits between `.md` and the section mark. Omitting it here matched no real issue
-    # at all - caught by the first CI run of this file's tests, not by reading it.
-    if ($Body -match '30-slices\.md`?\s*§\s*W(?<n>\d+)\s*@\s*`?(?<sha>[0-9a-fA-F]{7,40})`?') {
-        return [pscustomobject]@{ Slice = [int]$Matches['n']; Sha = $Matches['sha'] }
-    }
-    $null
-}
-
-function Get-TrackerIssue {
-    param([string] $Repository)
-
-    $ghArgs = @('issue', 'list', '--state', 'all', '--limit', '200', '--json', 'number,title,state,body')
-    if ($Repository) { $ghArgs += @('-R', $Repository) }
-
-    try {
-        $json = & gh @ghArgs 2>$null
-        if ($LASTEXITCODE -ne 0) {
-            return [pscustomobject]@{ Issues = @(); Failure = (New-Failure -Reason 'GhUnavailable' -Detail "gh exited $LASTEXITCODE") }
-        }
-    } catch {
-        return [pscustomobject]@{ Issues = @(); Failure = (New-Failure -Reason 'GhUnavailable' -Detail $_.Exception.Message) }
-    }
-
-    if ([string]::IsNullOrWhiteSpace(($json -join ''))) {
-        return [pscustomobject]@{ Issues = @(); Failure = (New-Failure -Reason 'GhUnavailable' -Detail 'gh returned no output') }
-    }
-
-    try {
-        $parsed = ($json -join "`n") | ConvertFrom-Json
-    } catch {
-        return [pscustomobject]@{ Issues = @(); Failure = (New-Failure -Reason 'TrackerUnreadable' -Detail $_.Exception.Message) }
-    }
-
-    [pscustomobject]@{ Issues = @($parsed); Failure = $null }
-}
-
-<#
-    Three outcomes, not two. An unknown object - a shallow clone, or a commit that was never
-    fetched - is not evidence the pin is stale, and folding it into NotAncestor would invent a
-    finding. It is the absence of an answer, which I12 requires be reported as such.
-#>
-function Test-CommitIsAncestor {
-    param([Parameter(Mandatory)][string] $Sha)
-
-    try {
-        & git merge-base --is-ancestor $Sha HEAD 2>$null | Out-Null
-        switch ($LASTEXITCODE) {
-            0       { 'Ancestor' }
-            1       { 'NotAncestor' }
-            default { 'Unresolvable' }
-        }
-    } catch {
-        'Unresolvable'
-    }
-}
-
-function Invoke-DriftCheck {
-    param([string] $SlicesPath, [string] $Repository)
-
-    $findings = [System.Collections.Generic.List[object]]::new()
-    $failures = [System.Collections.Generic.List[object]]::new()
-
-    $doc = Get-SliceCriteria -Path $SlicesPath
-    if ($doc.Failure) {
-        $failures.Add($doc.Failure)
-        return New-DriftResult -State 'NotEvaluated' -Failures $failures
-    }
-
-    if ($doc.Slices.ContainsKey(-1)) {
-        foreach ($stray in $doc.Slices[-1]) {
-            $failures.Add((New-Failure -Reason 'UnparseableCriterion' -Detail $stray))
-        }
-        $doc.Slices.Remove(-1)
-    }
-
-    $tracker = Get-TrackerIssue -Repository $Repository
-    if ($tracker.Failure) {
-        $failures.Add($tracker.Failure)
-        return New-DriftResult -State 'NotEvaluated' -Failures $failures
-    }
-
-    $landed = @($doc.Landed)
-    $unissuedLanded = [System.Collections.Generic.List[int]]::new()
-    $compared = 0
-
-    foreach ($number in ($doc.Slices.Keys | Sort-Object)) {
-        $docIds = @($doc.Slices[$number] | Sort-Object -Unique)
-        $issue  = $tracker.Issues | Where-Object { $_.title -match "^W$number\b" } | Select-Object -First 1
-
-        if (-not $issue) {
-            # A landed slice is not synced at all (the kit's skills/track/SKILL.md, "What syncs"),
-            # so there is no issue for /agentkit:track to open and this is not a finding. The ones
-            # here predate the tracker; reporting them made the gate permanently red, which
-            # is how three real findings sat unread.
-            if ($number -in $landed) {
-                $unissuedLanded.Add($number)
-                continue
-            }
-            $findings.Add((New-Finding -Kind 'NoIssue' -Slice "W$number" -Detail 'slice has no issue; /agentkit:track opens one' -Issue 0))
-            continue
-        }
-
-        $compared++
-        $issueIds = @(Get-IssueCriteria -Body $issue.body | Sort-Object -Unique)
-
-        foreach ($id in ($docIds | Where-Object { $_ -notin $issueIds })) {
-            $findings.Add((New-Finding -Kind 'InDocNotIssue' -Slice "W$number" -Detail $id -Issue $issue.number))
-        }
-        foreach ($id in ($issueIds | Where-Object { $_ -notin $docIds })) {
-            $findings.Add((New-Finding -Kind 'InIssueNotDoc' -Slice "W$number" -Detail $id -Issue $issue.number))
-        }
-    }
-
-    # In the kit's generic layout, a landed slice's body is retired once its issue closes,
-    # so only its pin can be checked. This repository does not retire slice bodies (every
-    # `### [x] W<n>` keeps its full "Done when" list - design/30-slices.md, "How this
-    # document is kept"), so a landed slice that still has an issue IS compared on ids above,
-    # and being landed only suppresses the NoIssue report for one that never had one.
-    foreach ($issue in $tracker.Issues) {
-        $pin = Get-IssuePin -Body $issue.body
-        if (-not $pin) { continue }
-
-        switch (Test-CommitIsAncestor -Sha $pin.Sha) {
-            'NotAncestor'  { $findings.Add((New-Finding -Kind 'PinNotAncestor' -Slice "W$($pin.Slice)" -Detail $pin.Sha -Issue $issue.number)) }
-            'Unresolvable' { $failures.Add((New-Failure -Reason 'PinUnresolvable' -Detail "#$($issue.number) pins $($pin.Sha), which this clone cannot resolve")) }
-        }
-    }
-
-    if ($unissuedLanded.Count -gt 0) {
-        Write-Verbose "Landed slices with no issue, not reported: $($unissuedLanded -join ', ')"
-    }
-
-    $state = if ($failures.Count -gt 0) { 'NotEvaluated' }
-             elseif ($findings.Count -gt 0) { 'Drifted' }
-             else { 'Clean' }
-
-    New-DriftResult -State $state -Findings $findings -Failures $failures -SlicesCompared $compared
-}
-
-function Get-DriftExitCode {
-    param([string] $State)
-    switch ($State) {
-        'Clean'        { 0 }
-        'Drifted'      { 1 }
-        'NotEvaluated' { 2 }
-        default        { throw "Unknown drift state: $State" }
-    }
-}
-
-function Write-DriftReport {
-    param([Parameter(Mandatory)][object] $Result)
-
-    Write-Host "Slices compared: $($Result.SlicesCompared)"
-    Write-Host "Findings: $($Result.Findings.Count)    Could not evaluate: $($Result.Failures.Count)"
-
-    foreach ($f in $Result.Findings) {
-        Write-Host "  [$($f.Kind)] $($f.Slice) $($f.Detail)$(if ($f.Issue) { " (#$($f.Issue))" })"
-    }
-    foreach ($f in $Result.Failures) {
-        Write-Host "  [$($f.Reason)] $($f.Detail)"
-    }
-
-    switch ($Result.State) {
-        'Clean'        { Write-Host 'No drift.' }
-        'Drifted'      { Write-Host 'Drift found. Which side is wrong is not this script''s call.' }
-        'NotEvaluated' { Write-Host 'Incomplete: at least one comparison did not run. This is NOT a clean result.' }
-    }
-}
-
-# Guards the exit-calling wrapper so this script's tests can dot-source it instead - that
-# defines every function above in the caller's scope, lets Mock intercept `gh` and `git`, and
-# skips straight past this block rather than exiting the test runner's own process.
-if ($MyInvocation.InvocationName -ne '.') {
-    if (-not $SlicesPath) {
-        $SlicesPath = Join-Path (Split-Path -Parent $PSScriptRoot) 'design/30-slices.md'
-    }
-    $result = Invoke-DriftCheck -SlicesPath $SlicesPath -Repository $Repository
-    if (-not $Quiet) { Write-DriftReport -Result $result }
-    $result
-    exit (Get-DriftExitCode -State $result.State)
-}