diff --git a/.claude/skills/alignment/SKILL.md b/.claude/skills/alignment/SKILL.md
index 8abb28d..2d78ac2 100644
--- a/.claude/skills/alignment/SKILL.md
+++ b/.claude/skills/alignment/SKILL.md
@@ -54,6 +54,16 @@ Review all changes in scope. If changes span multiple patterns below, apply all
      only within their module (including `as const` arrays that exist to derive a type) stay
      unexported — `export type` the type, not its source array. Verify with Grep before
      exporting "for future use"; the future caller can add the export.
+   - **Smell baseline** (Fowler, _Refactoring_ ch. 3 — judgement calls, never hard
+     violations; a documented project rule overrides, and skip anything eslint/Prettier
+     already enforces): mysterious name (rename — if no honest name comes, the design is
+     murky); data clumps (the same few params traveling together → one type); primitive
+     obsession (a string/number standing in for a domain concept); feature envy (a function
+     reaching into another module's data more than its own); repeated switches (the same
+     discriminant cascade at multiple sites → one shared map); message chains
+     (`a.b().c().d()` → hide the walk behind the first object); middle man (a layer that
+     only delegates → call the target directly). Duplication and speculative generality are
+     covered by the cleanup rules above.
 
 5. **Styles and UI**: When altering CSS or layout, follow or generalize existing patterns
    (CSS Modules + design tokens in `styles/tokens.css`) rather than writing from scratch. Don't
diff --git a/.claude/skills/eng-council/SKILL.md b/.claude/skills/eng-council/SKILL.md
index bfd5027..cc46ce7 100644
--- a/.claude/skills/eng-council/SKILL.md
+++ b/.claude/skills/eng-council/SKILL.md
@@ -126,7 +126,9 @@ that prevents data loss, or accessibility.
 **If it must be built — what shape?** Line up the existing instances of the kind (modal,
 store, service, hook, persistence path), name the canonical shape, list what to reuse, and
 flag what the new work might make deletable. An abstraction is earned only by ≥ 2 call
-sites that would use it today (deletion rule 3).
+sites that would use it today (deletion rule 3). The deletion test settles suspected
+pass-throughs: imagine the module deleted — if the complexity just vanishes, it was a
+shallow wrapper; only if it reappears across its callers was it earning its keep.
 
 No report scaffolding — these two answers are the output.
 
diff --git a/AGENTS.md b/AGENTS.md
index ac0d39b..fe2c17f 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -123,6 +123,12 @@ npm run format     # Prettier
   leads. When in doubt about existing behavior, read the spec. When you ship user-facing
   behavior, update the matching spec section in the same session; if spec and app
   disagree, the spec is stale — rewrite it deliberately, never drift silently.
+- **Reproduce before theorizing** — on a nontrivial bug, first build a command that goes
+  red on the exact symptom (failing test, script, headless-browser driver) and is fast,
+  deterministic, and runnable unattended; minimize the repro, then hypothesize against it.
+  Reading code to build a theory before that command exists is the failure mode. Write the
+  regression test before the fix; tag temporary debug logs with a unique prefix
+  (e.g. `[DEBUG-x7]`) so cleanup is one grep.
 - **Core-first** — for each feature, build the pure `src/core/` logic with tests before UI.
 - **Session wrap-up** — when the user signals the session is wrapping, run the review
   pass before any commit: `/doc-update` in-session first (flush unrecorded rationale),
@@ -172,6 +178,11 @@ carries — platform branches, state transitions, config-path writes, render ser
 not the strings it renders; if that logic is worth guarding, lift it into core/stores and
 test it there.
 
+Expected values come from an independent source of truth — a known-good literal, a worked
+example, the spec — never recomputed the way the implementation computes them. A
+tautological assertion (`expect(add(a, b)).toBe(a + b)`) passes by construction and can
+never disagree with the code.
+
 Component tests (happy-dom) share a harness shape: `createRoot` + `act` with
 `IS_REACT_ACT_ENVIRONMENT = true` set at module level, stores reset in `beforeEach`, and
 `vi.mock('../services/chart-renderer', …)` for anything that embeds a chart (vega-embed is
diff --git a/docs/exploration/skills-repos-review.md b/docs/exploration/skills-repos-review.md
new file mode 100644
index 0000000..a5a80e0
--- /dev/null
+++ b/docs/exploration/skills-repos-review.md
@@ -0,0 +1,73 @@
+# External skills repos review — mattpocock/skills and github/spec-kit
+
+_Point-in-time record, 2026-07-04. Both repos are shallow-cloned under
+`/Users/oleh/code/reference/` (`mattpocock-skills/`, `spec-kit/`) for grepping._
+
+## Question
+
+Do the general-purpose engineering-robustness skill sets — mattpocock/skills and
+github/spec-kit — contain anything our project skills (`/alignment`, `/eng-council`,
+`/council`, `/doc-update`) should adopt? Constraint: fold single additive ideas into
+existing skills; never install a parallel framework.
+
+## Verdict
+
+Neither repo is worth adopting wholesale. Our set covers the same ground with more rigor
+because it is project-specific and evidence-grounded where theirs is generic. Four
+discrete ideas were folded in (all from mattpocock/skills or shared with spec-kit);
+everything else was already covered or rejected.
+
+## The three approaches
+
+- **Ours** — a closed-loop system: `/alignment` enforces numbered project-specific checks
+  per diff; `/eng-council` and `/council` review from altitude under a hard evidence
+  requirement (file:line, tool output, cited canon); recurring findings promote into
+  architecture rules and new alignment checks. Built around the code-leads /
+  descriptive-spec regime and a deletion bias (LOC delta per finding).
+- **mattpocock/skills** — small composable per-task disciplines: grilling (relentless
+  one-question-at-a-time plan interviews), TDD, a bug-diagnosis loop, two-axis code review
+  (repo standards + Fowler smell baseline vs. originating spec), domain modeling
+  (`CONTEXT.md` glossary + ADRs), deep-module design vocabulary (Ousterhout/Feathers).
+  Same anti-framework philosophy as ours; its README positions against spec-kit explicitly.
+- **github/spec-kit** — a heavyweight spec-first pipeline (constitution → specify →
+  clarify → plan → tasks → analyze → implement → converge), seven-plus artifacts per
+  feature, gates between phases. Its thesis — the spec is the primary artifact, code its
+  expression — is the inverse of our regime, and it assumes greenfield feature branches,
+  team role separation, and business-stakeholder specs. Even `converge`, its only
+  code-vs-artifacts mode, treats undocumented code behavior as scope creep to justify or
+  remove, where our regime says the code is right and the spec gets rewritten.
+
+## Folded in (2026-07-04)
+
+| Idea                                                                                                                                                                        | Source            | Landed in                         |
+| --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------- | --------------------------------- |
+| Reproduce-before-theorizing debugging discipline: red-capable repro command before any hypothesis; minimize; regression test before fix; prefix-tagged debug logs           | `diagnosing-bugs` | AGENTS.md → AI Developer Protocol |
+| Fowler smell baseline as judgement-call heuristics (mysterious name, data clumps, primitive obsession, feature envy, repeated switches, message chains, middle man)         | `code-review`     | `/alignment` rule 4               |
+| Tautological-test rule: expected values from an independent source of truth, never recomputed the implementation's way                                                      | `tdd`             | AGENTS.md → Testing Philosophy    |
+| The deletion test for suspected pass-throughs: delete the module mentally — complexity vanishing means shallow wrapper, reappearing across callers means it earned its keep | `codebase-design` | `/eng-council` consult mode       |
+
+## Considered and rejected
+
+- **Grilling as a skill** — sessions here are already interactively driven, and the
+  harness's question tool plus explore-instead-of-ask covers the discipline. No standing
+  gap.
+- **`CONTEXT.md` glossary + ADRs (domain modeling)** — `docs/architecture/` +
+  `/doc-update` fill the same role with a stricter altitude bar; a second
+  decision-record home would split the record.
+- **spec-kit's constitution** — SOUL.md + the architecture playbook already are the
+  constitution, and ours is enforced mechanically (alignment checks), not re-read per
+  phase.
+- **spec-kit's "unit tests for English" checklists** (items test requirement quality:
+  completeness/clarity/measurability, banned Verify/Test verbs) — the standout idea of
+  the repo, but it targets prescriptive specs. Our spec is descriptive; its quality bar
+  is "matches the code", which alignment's spec-tracking check already enforces.
+- **spec-kit's bidirectional coverage / gap-type taxonomy** (`missing`/`partial`/
+  `contradicts`/`unrequested`) — both directions of spec↔code drift are already covered
+  by alignment's spec-tracking check and the eng-council Documentation seat.
+- **spec-kit's clarify mechanics** (fixed ambiguity taxonomy, Impact×Uncertainty question
+  cap, recommend-before-asking) — recommend-before-asking is already harness convention;
+  the rest is ceremony sized for teams, not a solo interactive loop.
+- **`improve-codebase-architecture` / HTML report** — `/eng-council` sweep covers it with
+  real evidence tooling (madge/knip/jscpd) and the metrics trend.
+- **`research`, `prototype`, `handoff`** — already covered by reference clones, the
+  one-off HTML showcase habit, and harness context management respectively.