Evidence: cold-agent trial, stock Spec Kit + Clewseau bundle delivers AC-HOME-15 end to end

.cursor/rules/cannon-eng02-context.mdc

@@ -0,0 +1,40 @@
+---
+description: What Potato Cannon / ENG-02 left in this repo, and how to read it
+globs: specs/002-smoke-build-info/**,.potato/**,.claude/**
+alwaysApply: false
+---
+
+# What Cannon is doing to this repo
+
+If you see `specs/002-smoke-build-info/`, a `.potato/` folder, or branches named `potato/HOM-*`, that is not stray product work — it is evidence of ENG-02, run against HomesFlow on purpose. Do not mistake it for real product scope.
+
+## Short version
+
+**Potato Cannon** (`potato-cannon` repo) is a kanban + agent orchestrator that runs Claude sessions through workflow phases with human approval gates. Rik uses it as the delivery mechanism for HomesFlow tickets going forward.
+
+**Golden Thread** is HomesFlow's layer on Spec-Kit: immutable IDs (`FR-`/`AC-` per `traceability.md` §3) and Gate 1 / Gate 2 (`traceability.md` §6). Spec-Kit itself does not know about it.
+
+**ENG-02** (spudnik) hypothesizes Cannon can drive HomesFlow's *real* installed spec-kit (`.claude/skills/speckit-*`) plus Golden Thread — not a hand-rolled approximation — with a kanban view and approval gates.
+
+## Reference example: HOM-2
+
+HOM-2 ("Stand up the Walking Skeleton") was the first end-to-end run of the `homesflow-speckit` template. Deliberately trivial (`FR-SMOKE-01`, `SMOKE` domain) — *not* real product scope, exempted from PRD-sourcing for that ticket only. Done. Read `specs/002-smoke-build-info/` for what a Cannon-driven ticket produces.
+
+Notable from that run: real skills invoked (not reimplemented); Gate 1 caught `plan.md` self-certifying IDs missing from `spec.md`; a mid-run `chat_ask` was answered via Cannon's dashboard and the session resumed.
+
+## Design decisions (Cannon side)
+
+- Phase gates only where `traceability.md` §6 says work stops (before/after implement). Specify → Clarify → Plan → Tasks → Analyze = one continuous resumed session.
+- No adversarial retry loops (`ralphLoop`/`taskLoop`). Real `speckit-implement` / `speckit-converge` already own that judgment.
+- Promote/demote resumes the Claude session with a digest of new comments/Q&A/artifacts — not a full context rebuild.
+
+## Still open
+
+**Thread Cannon** (fork Cannon into a Spec-Kit + Golden Thread–specific tool) was deferred until the walking skeleton finished; that decision is live again, not settled.
+
+## Go deeper
+
+- `potato-cannon` + its CLAUDE.md
+- `potato-cannon/apps/daemon/templates/workflows/homesflow-speckit/`
+- spudnik `OBJECTIVES.md` (ENG-02), `docs/field-notes-pipeline.md` (FN-05)
+- This repo's `traceability.md`

.cursor/rules/newbie-homesflow.mdc

@@ -0,0 +1,64 @@
+---
+description: HomesFlow trial engineer orientation — Spec Kit, Golden Thread, craft/PR hygiene
+alwaysApply: true
+---
+
+# Context Primer — HomesFlow trial engineer
+
+You are joining HomesFlow as a trial collaborator. Your job is to deliver a thin slice the HomesFlow way: Spec Kit discipline, Golden Thread integrity, then craft/PR hygiene. Do not try to match prior agent diffs or invent product scope.
+
+## How to work with Rik
+- High-signal, direct, verify before asserting.
+- Self-serve from the repo. Ask only when the PRD is silent or something is broken/off-repo.
+
+## Read first (in order)
+1. `CLAUDE.md` — entry + hard rules
+2. `.specify/memory/constitution.md` — process/architecture law (Hierarchy of Truth)
+3. `traceability.md` — Gate 2 mechanics
+4. `HomesFlow.prd.md` — **product scope only**
+5. `specs/001-mvp/dev-notes.md` — env + operational state
+6. `specs/001-mvp/craft-conventions.md` — craft + delivery
+
+## Hard rules
+- **Product scope** comes only from `HomesFlow.prd.md`. Do not invent, extrapolate, or assume requirements. If the PRD is silent, ask or propose a PRD change — do not ship guessed behavior.
+- **Process/architecture** comes from the constitution. Do not invent product behavior from the constitution, glossary, story map, or craft docs.
+- **Story map** (StoriesOnBoard) is a human planning aid only — not Gate-2-enforced. Promote into the PRD when it becomes product truth.
+- **Gate 2** proves the PRD is *true of the system* (IDs ↔ `@covers` ↔ `test_AC_*` or tracked debt). It does not prove the PRD is the right product.
+
+## Spec Kit loop (expected)
+For the assigned thin slice:
+1. Confirm the authoritative IDs in `HomesFlow.prd.md` (Rik amends the PRD first if needed).
+2. Fan into `specs/001-mvp/` (spec → plan → tasks) without minting new requirement IDs in the feature folder.
+3. Implement only what tasks authorize; keep `Traces:` on every task.
+4. Prove with `@covers` and `test_AC_*` (or leave unchecked tasks as tracked debt — never silent gaps).
+5. Run `bash scripts/check-traceability.sh` before you call the slice done.
+
+## Golden Thread (commands)
+```bash
+bash scripts/check-traceability.sh          # must PASS
+bash scripts/check-traceability.sh --json   # per-ID detail
+```
+Coverage matrix (`coverage.md`) is a portfolio snapshot; CI enforces the gate script, not file freshness.
+
+## Craft / PR hygiene
+```text
+feature branch → PR → craft-gate (+ Sonar) green → merge to main
+```
+- Branch naming for cloud agents: `cursor/<descriptive-name>-7716`
+- Fill `.github/pull_request_template.md` (task IDs, Traces, Gate 2 evidence)
+- Do not push product/craft changes straight to `main`
+- Follow `craft-conventions.md` (SwiftLint, policy objects where logic is testable)
+
+## Your trial assignment
+Rik will give you **one micro acceptance criterion** newly added to the PRD (atomic, unit-testable). Take it through Spec Kit → implementation → Gate 2 green → PR.
+
+**Success looks like:** clean Spec Kit artifacts, unbroken golden thread for that ID, and a reviewable PR.
+**Non-goals:** matching a prior check-in’s code; rebuilding the thread visualizer; expanding scope “while you’re in there.”
+
+## Local assumptions (confirm if relevant)
+- Supabase local: `127.0.0.1:54321`; `Secrets.xcconfig` from `.example`
+- Xcode: `cd ios && xcodegen generate`
+- Paths: `ios/HomesFlow/`, `ios/HomesFlowTests/`
+
+## Starting the session
+Confirm: (1) you have the micro-AC ID and PRD text, (2) local vs cloud env, (3) any blockers not in `dev-notes`. Then proceed.

.specify/extensions.yml

@@ -1,23 +1,14 @@
 installed:
-- agent-context
+- clewseau-gate
 settings:
   auto_execute_hooks: true
 hooks:
-  after_specify:
-  - extension: agent-context
-    command: speckit.agent-context.update
+  after_implement:
+  - extension: clewseau-gate
+    command: speckit.clewseau-gate.check
     enabled: true
     optional: true
     priority: 10
-    prompt: Execute speckit.agent-context.update?
-    description: Refresh agent context after specification
-    condition: null
-  after_plan:
-  - extension: agent-context
-    command: speckit.agent-context.update
-    enabled: true
-    optional: true
-    priority: 10
-    prompt: Execute speckit.agent-context.update?
-    description: Refresh agent context after planning
+    prompt: Run Clewseau Gate 2 traceability check?
+    description: Refuse silent gaps after implementation; refresh clew.json
     condition: null

.specify/extensions/.registry

@@ -1,19 +1,19 @@
 {
   "schema_version": "1.0",
   "extensions": {
-    "agent-context": {
-      "version": "1.0.0",
+    "clewseau-gate": {
+      "version": "0.2.0",
       "source": "local",
-      "manifest_hash": "sha256:9a1dc02d2d0139bb03860392ecacef79183be2c442feda2f9ccaa4e5907b1e47",
+      "manifest_hash": "sha256:c540ad473eb810d0952c51812571222b0d45b9f7aa5828d252d0cb5809110c89",
       "enabled": true,
       "priority": 10,
       "registered_commands": {
         "claude": [
-          "speckit.agent-context.update"
+          "speckit.clewseau-gate.check"
         ]
       },
       "registered_skills": [],
-      "installed_at": "2026-06-26T14:27:06.775090+00:00"
+      "installed_at": "2026-08-06T20:51:47.837406+00:00"
     }
   }
 }