@ahmed-g-gad/apothem 0.1.1

package/src/apothem/commands/README.md

@@ -0,0 +1,125 @@
+<!-- SPDX-License-Identifier: MIT -->
+
+# Commands
+
+Slash-command definitions — flat `.md` files, each defining one `/command-name` invocable. A command file's YAML frontmatter declares the command's identity and argument surface; its body is the multi-step workflow the command runs. Forty-five commands register at the top level — the seven **planning-pipeline** stage commands plus the `/plan-amend` amendment command and the `/plan` wrapped-workflow orchestrator, the thirteen **research-pipeline** stage commands plus the `/research` wrapped-workflow orchestrator, the eleven **audit / review** passes plus the `/audit` wrapped-workflow orchestrator and the `/fortress` closed-loop hardening orchestrator, three **cohort** commands (`test-suite`, `eval`, `release-readiness`), four **deployment / elevation** commands (`freshify`, `github-deploy-fresh`, `github-deploy-next`, `elevate`), and two **operator-workflow** commands (`workflow`, `projectify`). Each planning and research stage is an independently invocable `/plan-<stage>` / `/research-<stage>` command, and the eleven audit passes are each individually invocable.
+
+## Plan pipeline — seven first-class `/plan-<stage>` commands
+
+The end-to-end planning workflow — prose → spec → plan suite → review → conditional architecture design → execution → status — is seven first-class commands, one per stage. Each is independently invocable as `/plan-<stage>`, preserving every per-stage gate and structured-inquiry discipline, with the three-tier scalability framework (small / medium / large, via [`../lib/plan_tiers.py`](../lib/plan_tiers.py)) governing suites up to thousands of tasks. Each non-initial stage carries a `## Sequence Gate` that refuses out-of-order invocation (absent `--override`, recorded as a finding) with a definitive `Blocked: run /plan-<predecessor> first` message. `/plan-design` runs only when `/plan-review` classifies the suite as architecture-bearing; non-architecture suites proceed directly to `/plan-execute`. `/plan-status` is orthogonal read-only at any point. `/plan-amend` amends an existing suite without destroying prior resolved decisions, re-deriving only affected downstream artifacts.
+
+Each command is a flat top-level command file, registered by the non-recursive `commands/*.md` glob like every other command:
+
+| Command | File | Purpose |
+|---------|------|---------|
+| `/plan-spec` | [`plan-spec.md`](plan-spec.md) | Refine free-form prose / raw notes into a spec-grade `_spec/spec.md`. Six transformation phases under four operational disciplines; every ambiguity surfaces via the structured-inquiry channel. `--quick` writes a lightweight project-local plan file instead. |
+| `/plan-generate` | [`plan-generate.md`](plan-generate.md) | Generate a complete Master Plan Suite from the user's raw prose / requirements. |
+| `/plan-review` | [`plan-review.md`](plan-review.md) | Forensic audit of an existing plan suite — prose fidelity, internal consistency, quality / gap analysis — then refine through interactive Q&A. |
+| `/plan-design` | [`plan-design.md`](plan-design.md) | Conditional architecture gate. Produces `_inputs/design.md` for architecture-bearing suites after review and before execute; skipped explicitly for non-architecture suites. |
+| `/plan-audit` | [`plan-audit.md`](plan-audit.md) | Closed-loop pipeline audit and active remediation — brings a suite to zero open findings, writes bounded reports under `_outputs/`, routes residual work to `*-maintenance`. |
+| `/plan-execute` | [`plan-execute.md`](plan-execute.md) | Execute a specific phase from a Master Plan Suite with conformity checking and quality gates. |
+| `/plan-status` | [`plan-status.md`](plan-status.md) | Read-only plan-suite progress report — status against task / phase / artifact dimensions, no modification. |
+| `/plan-amend` | [`plan-amend.md`](plan-amend.md) | Amend / extend / refine / revert / weave an existing plan suite without destroying prior resolved decisions; re-derives only affected downstream artifacts and recommends the affected downstream stage (`/plan-review`). |
+| `/plan` | [`plan.md`](plan.md) | The plan pipeline wrapped as a single dynamic multi-agent workflow (not itself a stage) — drives a planning mission from raw prose to executed phases by dispatching the first-class stages (`plan-spec → plan-generate → plan-review → plan-design` conditional `→ plan-execute`) as workflow phases under named Handoff-Manifest return contracts, routing each stage hand-off through a refute-by-default verification pass, and emitting a deterministic result with a single recommended next move. Stage logic stays first-class in `plan-*.md`; chaining + dispatch are opt-in via `--autonomous`, halting at each stage boundary by default. |
+
+The top-level `command_skills` propagation registers every planning-pipeline command (the seven stages, `/plan-amend`, and the `/plan` wrapped-workflow orchestrator) and the audit / review commands alike by the non-recursive `commands/*.md` glob.
+
+## Research pipeline — thirteen first-class `/research-<stage>` commands
+
+The end-to-end research workflow — idea → question → theory → sources → synthesis → proposal → study design → experiment → analysis → paper → peer review → publish → disseminate — is thirteen first-class commands, one per stage, mirroring the plan pipeline. Each is independently invocable as `/research-<stage>`, with a `## Sequence Gate` that refuses out-of-order invocation (absent `--override`, recorded as a finding) with a definitive `Blocked: run /research-<predecessor> first` message, and a Handoff Manifest chaining each stage to the next. The pipeline operates under ten rigor mandates (authoritative sources · reproducibility · falsifiability · citation integrity · preregistration · ethics / conflicts · statistical rigor · open-science / FAIR · reporting-guideline conformance · theoretical grounding / impact) and resolves the [`research-suite`](../skills/research-suite/) knowledge surface by path.
+
+| Command | File | Purpose |
+|---------|------|---------|
+| `/research-ideate` | [`research-ideate.md`](research-ideate.md) | Zero-ideation entry — formulate the problem space from a domain seed, generate and rank candidate research questions, and frame the opportunity the spec stage consumes. |
+| `/research-spec` | [`research-spec.md`](research-spec.md) | Frame a candidate question into a spec-grade research spec — falsifiable hypotheses, scope, inclusion / exclusion criteria, success metrics. `--quick` writes a lightweight project-local research brief. |
+| `/research-theory` | [`research-theory.md`](research-theory.md) | Build the foundational conceptual framework and theory-of-change — name the constructs, their relationships, and the mechanism the study tests. |
+| `/research-sources` | [`research-sources.md`](research-sources.md) | Systematic source collection — decompose into sub-queries, parallel discovery via `research-scout` + `multi-source-research`, dedup, rank by authority / recency / relevance, screen against inclusion criteria, extract per source. |
+| `/research-synthesis` | [`research-synthesis.md`](research-synthesis.md) | State-of-the-art map + literature matrix + explicit gap statement; every load-bearing claim adversarially verified by `fact-checker` via the `source-synthesis` skill. |
+| `/research-proposal` | [`research-proposal.md`](research-proposal.md) | Objectives / aims, feasibility assessment, impact pathway, and the preregistration plan — the fundable proposal the study design operationalizes. |
+| `/research-design` | [`research-design.md`](research-design.md) | Operationalize hypotheses into testable predictions; design variables / controls / sample / instruments / power analysis / threats-to-validity; freeze the analysis plan as a preregistration. |
+| `/research-experiment` | [`research-experiment.md`](research-experiment.md) | Run the designed study and capture raw data with full provenance + a reproducibility manifest (environment, seed, protocol, version pins). |
+| `/research-analysis` | [`research-analysis.md`](research-analysis.md) | Analyze per the preregistered plan — effect sizes + confidence intervals, robustness / sensitivity checks, disclosed deviations, figures / tables. |
+| `/research-paper` | [`research-paper.md`](research-paper.md) | Assemble a top-tier paper draft (abstract → conclusion + references) from synthesis + design + analysis; every citation verified to resolve to a real source. |
+| `/research-review` | [`research-review.md`](research-review.md) | Peer-review-grade adversarial critique — reviewer scorecard (novelty / rigor / reproducibility / clarity / ethics) + a severity-triaged required-revision list. |
+| `/research-publish` | [`research-publish.md`](research-publish.md) | Venue-formatted submission package — supplementary materials, data / code-availability statement, cover letter, ethics / COI declarations, preprint / DOI plan, submission checklist. |
+| `/research-disseminate` | [`research-disseminate.md`](research-disseminate.md) | Post-acceptance dissemination, impact, and archival — plain-language summary, channel plan, FAIR data / code deposit, persistent-identifier registration, and the impact-tracking record. |
+| `/research` | [`research.md`](research.md) | The research pipeline wrapped as a single dynamic multi-agent workflow (not itself a stage) — drives a research mission from a raw idea to a disseminated, peer-reviewed paper by dispatching the stages as workflow phases under named Handoff-Manifest return contracts, routing each hand-off through a refute-by-default verification pass. Chaining is opt-in via `--autonomous`, halting at each stage boundary by default; `--quick` runs `research-sources` + `research-synthesis` only for a single-shot cited report. |
+
+## Audit / review passes
+
+Operator-driven review passes against a deployed repository. Each walks a defined surface and emits a severity-triaged (HIGH / MEDIUM / LOW) findings artifact under the consuming suite's `_inputs/` directory with concrete-driver rationale per finding. The [`/audit`](audit.md) **wrapped-workflow orchestrator** drives all eleven as a single parallel sweep — the audit-fortress analogue of `/plan` and `/research` — fanning the dimensions out under named findings return contracts, refute-by-default-verifying each finding, and synthesizing one deduplicated, severity-triaged report. `/audit` is report-only (remediation routes to `/elevate` or the owning surface); each pass below is also individually invocable.
+
+The [`/fortress`](fortress.md) **closed-loop hardening orchestrator** closes the loop report-only `/audit` opens: it detects through `/audit`, adversarially verifies each finding, remediates every survivor at its owning surface, re-audits in a bounded loop until the walls hold, and gates the result through `/release-readiness` — the production-hardening analogue of `/plan` and `/research`. Distinct from `/elevate` (broad open-loop whole-repo SOTA lift) and `/release-readiness` (single READY / BLOCKED verdict), `/fortress` is the security/production-scoped detect → remediate → re-audit → gate loop.
+
+| Command | Surface audited |
+|---------|-----------------|
+| [`code-review.md`](code-review.md) | Per-file code-quality review — readability, naming, complexity, magic numbers, comment quality, against the four code-craft rules and the ten quality dimensions. |
+| [`code-audit.md`](code-audit.md) | Cross-file forensic code audit — hidden coupling, layer-boundary violations, type-hint accuracy, coverage gaps, dead code, duplicates. The cross-file counterpart to `/code-review`. |
+| [`architecture-review.md`](architecture-review.md) | Architectural-integrity review against `_inputs/design.md` and the clean-architecture layer discipline. |
+| [`docs-review.md`](docs-review.md) | Documentation review against the Markdown code-craft and ten-dimension rules — prose clarity, link integrity, citation completeness, public-API coverage. |
+| [`security-audit.md`](security-audit.md) | Security posture audit against OWASP ASVS, OWASP Top 10, and CWE Top 25 — secrets, injection surfaces, deserialization, path-traversal, dependency CVEs. |
+| [`dependency-audit.md`](dependency-audit.md) | Per-dependency audit — license compatibility, CVE status, deprecation, pinned-vs-range posture, transitive depth. |
+| [`supply-chain-audit.md`](supply-chain-audit.md) | Supply-chain audit against SLSA + Sigstore + SBOM standards — provenance, signing, SBOM completeness, action pinning. |
+| [`threat-model-audit.md`](threat-model-audit.md) | Threat-modeling audit against STRIDE + PASTA — trust boundaries, threat actors, mitigation posture, residual risk. |
+| [`perf-audit.md`](perf-audit.md) | Performance audit against the per-class budgets in `../rules/performance-discipline.md`, via the benchmark drivers under `../benchmarks/`. |
+| [`a11y-audit.md`](a11y-audit.md) | Accessibility audit against WCAG 2.2 AA — semantic HTML, ARIA, keyboard-navigability, contrast, alt-text. |
+| [`ux-review.md`](ux-review.md) | Developer-experience review — CLI ergonomics, installation flow, documentation discoverability — against clig.dev, the NN/g heuristics, and GNU coreutils conventions. |
+
+## Cohort commands
+
+Operator-driven workflow commands, each orchestrating a multi-step engagement across its own skill / agent cohort. Output lands at the consuming suite's `_inputs/` directory with concrete-driver rationale per finding.
+
+| Command | Cohort | Purpose |
+|---------|--------|---------|
+| [`test-suite.md`](test-suite.md) | developer | Behavior-first test authoring and execution — discover the host's test framework, author behavior-shaped AAA tests, run them, triage failures, and report coverage gaps against critical paths. |
+| [`eval.md`](eval.md) | ai-engineering | Run a model-agnostic language-model evaluation campaign — define dataset and scorer, score every output, aggregate metrics with a per-category breakdown, and surface regressions against the prior baseline. |
+| [`release-readiness.md`](release-readiness.md) | developer / security | Pre-release gate sweep against the production-ready discipline — quality matrix, dependency risk, supply-chain checks, visibility surfaces, CHANGELOG currency, and version-to-tag consistency, emitting a single READY / BLOCKED verdict. |
+
+## Deployment / elevation
+
+Repository-wide freshening, release, and SOTA-elevation commands. Each routes
+every destructive step through the structured-inquiry confirmation channel with
+in-place freshening as the default.
+
+| Command | Purpose |
+|---------|---------|
+| [`freshify.md`](freshify.md) | Host- and forge-AGNOSTIC freshening core — purges caches and stale artifacts, removes legacy/obsolete narrative and back-references, normalizes file/folder naming and drives every surface to maximal naturalness and coherence, enforces a current-version-only facade, and drives the host's discovered gates to green. Specialized by `/github-deploy-fresh`. |
+| [`github-deploy-fresh.md`](github-deploy-fresh.md) | GitHub specialization of `/freshify` — a single fresh `release: <repo-name> v0.1.0` to `origin/main`, strictly-green (maximal-score where applicable) workflows, a curated first-version CHANGELOG, and a trace-free repository (delete+recreate available as a metadata-preserving, confirmation-gated `MAY`). |
+| [`github-deploy-next.md`](github-deploy-next.md) | The next-release-cycle sibling — merge PRs, resolve issues, SemVer bump, Keep-a-Changelog roll, tag + sign + publish (signing host-discovered), and current-version release notes; merge and publish are confirmation-gated. |
+| [`elevate.md`](elevate.md) | Aggressive, relentless, multi-agent, zero-trust, open-loop **master** repository elevation — a major multi-pass undertaking (not a quick amendment) that detects the technically-auditable nuances by dispatching the report-only `/audit` fortress (reusing its eleven passes, never re-deriving them), runs bespoke critique over the SOTA/elevation dimensions no audit covers, stops at each nuance folder-by-folder, file-by-file, line-by-line, remediates with the most appropriate action through the dimension-native skills (destruction included), propagates every change across the reference graph, and culminates trace-free via `/freshify`. |
+
+## Operator workflow
+
+General-purpose operator commands that harness a whole mission end to end. Each emits deterministic output closing on a single recommended next move; multi-agent dispatch and multi-step autonomy are opt-in / confirmation-gated, never default-on.
+
+| Command | Purpose |
+|---------|---------|
+| [`workflow.md`](workflow.md) | Workflow-harnessing command (entry form `/goal <<mission>>`) — decomposes a mission and drives genuinely-independent multi-agent dynamic workflows under named return contracts, subjects every finding to a refute-by-default verification pass before it survives, and self-augments from current authoritative sources, not memory alone. |
+| [`projectify.md`](projectify.md) | Chat-app Project elevation command — produces the three deliverables (Description, Instruction, knowledge Files) for a Claude Project / ChatGPT Custom GPT / Gemini Gem through a structured-inquiry-saturated elicitation, holding knowledge files within a measurable per-platform context budget. |
+
+## Frontmatter contract
+
+Command frontmatter is validated against [`../schemas/command.schema.json`](../schemas/command.schema.json). Observed fields:
+
+- `name` — command identifier; the slash command is `/<name>`.
+- `version` / `updated` — semantic version and ISO-8601 revision date.
+- `description` — statement of what the command does.
+- `argument-hint` — the command's argument / flag surface, shown in invocation help.
+
+The body after the frontmatter is the command's workflow specification: ordered steps, gates, structured-inquiry invocation points, and output contract.
+
+## Conventions
+
+- One flat `.md` file per command; filename stem equals the `name` field and the slash-command name.
+- Every file carries the canonical single-line SPDX license header.
+- Plan-pipeline commands are also surfaced as skills; the `/plan-<stage>` command family resolves the plan-suite template under [`../skills/plan-suite/`](../skills/plan-suite/) by path.
+- Harness installation of custom slash commands is controlled by each adapter's `capabilities.yml` `custom_command_support` value. `yes` means the adapter propagates this cohort to a native command directory; `no` means the surface is unsupported; `discovery-pending` means the adapter must not assume native support until its pin is refreshed.
+- Planning-technique discipline applied during these workflows is specified in `../rules/planning-techniques.md`.
+
+## Operating in this folder
+
+- **This folder is swept by the agnosticism matcher.** Command definitions stay harness-neutral: name no harness, model, or tool as privileged, and **do not pre-set an effort or model preference** in frontmatter or body. Native-support routing is a per-adapter `capabilities.yml` concern, not a per-command claim.
+- Command files carry determinism and recommend-next-step gates: a definitive forward-move block closes a workflow's terminal surface, and directive prose stays hedging-free.
+- **Adding a command:** author a flat `commands/<name>.md` file (stem = `name` = slash name) with schema-valid frontmatter and a workflow body. The discovery glob is non-recursive, so a command must sit at the top level (not nested in a subdirectory) to register. Surface ambiguity through the structured-inquiry channel or a `TODO(clarify)` marker — never invented.
+- A documented public command surface change updates its `site/content/docs/` page in the same change-set.
+- Validate with `python -m apothem.conformity.gate --all .` (name+description floor, determinism, recommend-next-step) and `python -m pytest`.

package/src/apothem/commands/a11y-audit.md

@@ -0,0 +1,203 @@
+---
+name: "a11y-audit"
+version: "0.1.0"
+updated: "2026-06-16"
+description: "Operator-driven accessibility audit pass against WCAG 2.2 AA. Walks every rendered page of a deployed web surface (documentation site, landing portal, in-app surfaces) via ax-core + Pa11y + Lighthouse Accessibility, attests each issue against the WCAG 2.2 success-criterion catalog (including the six 2.2-new criteria — 2.4.11 Focus Not Obscured, 2.5.7 Dragging Movements, 2.5.8 Target Size, 3.3.7 Redundant Entry, 3.3.8 Accessible Authentication, plus the carried-forward AA floor), and emits per-page findings — HIGH/MEDIUM/LOW severity-triaged with concrete-driver rationale per finding. Read-only diagnostics; never remediates. Output lands at the consuming suite's _inputs/a11y-audit-findings.md. Invoke with a site path or URL, or --focus PAGE_OR_DIR to audit a recent docs change-set incrementally."
+argument-hint: "[path/to/site/ or URL] [--focus PAGE_OR_DIR] [--dry-run]"
+disable-model-invocation: true
+portability: "universal"
+allowed-tools: "*"
+---
+
+<!-- SPDX-License-Identifier: MIT -->
+
+# /a11y-audit — Per-Page Accessibility Audit (WCAG 2.2 AA)
+
+---
+
+## Role
+
+You are the user's **Accessibility Engineer** and **Cognitive Insurgent** (`rules/cognitive-identity.md`), operating as **auditor-as-instrument-not-author**. This is a forensic surface: it surfaces WCAG 2.2 AA non-conformance, ARIA-pattern divergence, and keyboard-trap occurrences against the canonical accessibility standards — it never authors the fix.
+
+Apply the cognitive filters per `rules/cognitive-identity.md` §2 and attest the touched axs from the §1 seven-axs taxonomy. For accessibility triage, **Tooling and Observability are load-bearing** — ax-core / Pa11y / Lighthouse instrumentation, with the rendered DOM as the observed surface.
+
+---
+
+## Instructions
+
+Execute `/a11y-audit`: ingest the deployed web surface (local rendered site, staging URL, or production URL), walk every rendered page, apply WCAG 2.2 AA criteria via ax-core + Pa11y + Lighthouse Accessibility, and emit a per-page findings artifact at the consuming suite's `_inputs/a11y-audit-findings.md` ready for downstream remediation.
+
+Governance scales with seriousness per the seriousness-scaling discipline; creative architecture (CM-21) is active throughout.
+
+---
+
+## Pipeline Contract
+
+**Pipeline position.** Terminal review-fortress command at the a11y slot. It consumes the deployed web surface state — every reachable rendered page under the operator-supplied root — and emits read-only accessibility diagnostics for downstream remediation. It modifies no source.
+
+**Audit-fortress sequence.** Position **7 of 11**. **Upstream:** `/ux-review`. **Downstream:** `/docs-review`. Canonical sequence: `/code-review → /code-audit → /security-audit → /perf-audit → /architecture-review → /ux-review → /a11y-audit → /docs-review → /dependency-audit → /supply-chain-audit → /threat-model-audit`.
+
+**Handoff Manifest.**
+
+- **Consumed.** The deployed web surface (filesystem path to a rendered static site, staging URL, or production URL). No upstream manifest is required; the command operates against deployed state. When a Handoff Manifest exists at `_inputs/handoff-manifest.yml`, prior fortress attestations are read as context but do not gate execution.
+- **Emitted.** The findings artifact at `_inputs/a11y-audit-findings.md`, plus an optional Handoff Manifest augmentation carrying the per-page finding count, per-severity breakdown, per-WCAG-criterion attestation, per-axis seven-axs attestation, and the audit's `verified:` date.
+
+**Pre-flight inquiry.** Phase 0 emits the typed inquiry set per `rules/authority-inquiry.md` when the deployed-surface shape is ambiguous (the root URL returns no reachable pages; the focus argument points at a non-existent page; the rendered site requires authentication credentials the audit does not carry). Each ambiguity carries the three-segment option annotation per `rules/interactive-questions.md` §3.
+
+**Pre-emission gate.** Phase 4 runs the fifteen-bar pre-emission gate (`rules/pre-emission-gate.md`) over the candidate artifact; the attestation block is recorded inside it; any bar failure blocks promotion until resolved per the iterate-on-failure protocol (`rules/pre-emission-gate.md` §3).
+
+### Inquiry Cadence (D4)
+
+Operate at **maximal structured-inquiry saturation**. Every severity ratification, borderline WCAG-conformance call (e.g. contrast at 4.49:1 versus the 4.5:1 floor), axis-attestation gap, and gate-bar `n/a (with reason)` marking routes through the canonical channel (`rules/interactive-questions.md` §1) — free-form prose questions as primary input are forbidden. Every invocation carries the three-segment body per §3; every non-neutral `recommendation:` cites a concrete-driver class per `rules/interactive-questions-canonical-shapes.md` §3.2.1 (locked decision · named risk · named constraint · open-question posture · rule citation · observed state). Up to four questions batch per invocation. Question-fatigue-optimization is FORBIDDEN.
+
+---
+
+## Foundational Stanzas
+
+The four standing surfaces every operator inherits per the canonical project voice at `AGENTS.md` plus the active harness mirror.
+
+### Refusal & Escalation
+
+REFUSE any task exceeding this command's mission (the per-page accessibility findings artifact for a deployed web surface against WCAG 2.2 AA). Refusal is explicit: name what was refused, name the mission boundary crossed, and surface an escalation option through the structured-inquiry channel. REFUSE audit against a surface whose page set is genuinely unbounded without operator-supplied focus narrowing (the crawl must terminate). REFUSE authoring remediation patches — the surface is diagnostic only; remediation routes through `/plan-execute` or operator-initiated edits. REFUSE audit against a higher conformance level (WCAG 2.2 AAA) without explicit operator ratification — WCAG 2.2 AA is the canonical default.
+
+### Output Surface
+
+The findings artifact lands at the consuming suite's `_inputs/a11y-audit-findings.md` per the suite-locality invariant (`rules/context-management.md` §2.6.1). Plan-internal files are banner-exempt per the `.plans/**` class at `src/apothem/schemas/header-exceptions.txt`, so `scripts/inject-header.{sh,py}` is NOT invoked. NEVER write outside the suite folder; NEVER write to a global plans directory under any harness's config root from a downstream-project context; NEVER write to any other global-ecosystem location; NEVER modify any rendered page or underlying source.
+
+### File-Authoring Contract
+
+The findings artifact is banner-exempt per the `.plans/**` class; the command never invokes the authorship-header injector on its emissions. Every page/URL citation is documentary (`page:selector`); the underlying source file is never written.
+
+### Structured Inquiry on Ambiguity
+
+Route through the structured-inquiry channel with the three-segment annotation (`rules/interactive-questions.md` §3) on any uncertainty about page-set scope, focus boundary, borderline contrast / focus-visibility / heading-hierarchy severity, or multi-axis attestation. Free-form prose questions as primary input are forbidden. NEVER fabricate findings — every finding cites a concrete `page:selector` (or `page:line` for source-mapped findings), a WCAG 2.2 success criterion, and the detecting tool (ax-core rule ID, Pa11y issue ID, Lighthouse audit ID).
+
+---
+
+## Inputs
+
+| Argument | Type | Required | Description |
+| -------- | ---- | -------- | ----------- |
+| `path/to/site/ or URL` | Path or URL | Yes | Root of the deployed web surface — either a filesystem path to a rendered static site (e.g. `site/dist/` after the build) OR a reachable URL (staging or production). The command refuses when neither resolves. |
+| `--focus PAGE_OR_DIR` | Path or URL fragment | No | Restrict the per-page walk to a single page or subtree under the root. Path resolves relative to the root; URL fragment matches the leading path segment. Useful for auditing a recent docs change-set incrementally. |
+| `--dry-run` | Flag | No | Report what would be audited — no artifact emitted. Enumerates the reachable page count, the per-tool invocation plan, and any pre-flight inquiries that would fire. |
+
+---
+
+## Workflow — Five Audit Phases
+
+### Phase 0 — Input Ingest
+
+Read the deployed web surface in full. Deploy a Research Team (CM-25A) — one agent per top-level page-group (`/`, `/docs/`, `/blog/`, `/pricing/`, etc. as the host's discovered navigation surfaces). Each agent returns a structured page inventory ≤ 500 tokens (CM-25C), required fields `status` · `page-list` · `per-group-count` · `gaps`.
+
+**Required reads.**
+
+- The host's sitemap surface (`sitemap.xml`, `robots.txt`, `manifest.json`, or the host-discovered docs-site index) per `rules/host-discovery-manifests.md` §1 — every discovered navigation convention anchors the per-page audit bar.
+- Every reachable page under the root matching the focus narrowing (or every reachable page when no focus is supplied), capped at a host-discoverable crawl ceiling to ensure termination.
+
+**Externalize the inventory** at `_inputs/a11y-audit-inventory.md` (free-form `{kebab-case-topic}.md` per `rules/context-management-scratch.md` §1): reachable page count, per-group count, the host's ratified accessibility-tooling configuration (ax-core / Pa11y / Lighthouse config discovered at the manifest walk), and any `--focus` narrowing.
+
+### Phase 1 — Per-Page Audit
+
+Apply WCAG 2.2 AA per page via three complementary tools:
+
+- **ax-core CLI** — the canonical ax-core rule set per page. Each violation surfaces as a candidate with `rule-id`, `impact` (minor/moderate/serious/critical), node selector, and `help-url`.
+- **Pa11y** — the HTML_CodeSniffer engine plus ax-core (union of detected issues). Each issue surfaces with `code` (WCAG criterion ID), `type` (error/warning/notice), `selector`, `context` (HTML snippet), and `message`.
+- **Lighthouse Accessibility** — the Lighthouse Accessibility audit suite. Each failing audit surfaces with `audit-id`, `score`, `score-display-mode`, and `description`.
+
+Attest every detected issue against the WCAG 2.2 success-criterion catalog. The load-bearing AA criteria (non-exhaustive):
+
+- **1.1.1 Non-text Content** (alt-text presence and quality)
+- **1.3.1 Info and Relationships** (semantic HTML, ARIA roles)
+- **1.4.3 Contrast (Minimum)** (4.5:1 for normal text, 3:1 for large text)
+- **1.4.11 Non-text Contrast** (UI-component and graphical-object contrast at 3:1)
+- **2.1.1 Keyboard** (every interactive surface keyboard-operable)
+- **2.4.3 Focus Order** (predictable, meaningful tab order)
+- **2.4.7 Focus Visible** (every focusable element renders a visible focus indicator)
+- **2.4.11 Focus Not Obscured (Minimum)** — *new in 2.2 AA* (focused element fully or partially visible)
+- **2.5.7 Dragging Movements** — *new in 2.2 AA* (every drag operation has a single-pointer alternative)
+- **2.5.8 Target Size (Minimum)** — *new in 2.2 AA* (every interactive target ≥ 24×24 CSS px)
+- **3.3.7 Redundant Entry** — *new in 2.2 AA* (previously-entered information auto-populated or selectable)
+- **3.3.8 Accessible Authentication (Minimum)** — *new in 2.2 AA* (no cognitive-function test required unless an alternative exists)
+- **4.1.2 Name, Role, Value** (every UI component carries a programmatically determinable name, role, and state)
+
+**Externalize per-page drafts** at `_inputs/a11y-audit-per-page/` (one Markdown file per audited page), each enumerating raw findings with `page:selector` citations plus the detecting tool's rule ID before triage.
+
+### Phase 2 — Per-Finding Triage
+
+Assign severity from `{HIGH, MEDIUM, LOW}` with concrete-driver rationale (`rules/interactive-questions-canonical-shapes.md` §3.2.1):
+
+- **HIGH** — an a11y blocker (keyboard trap with no escape · missing alt-text on an informational image · contrast below 3:1 on body text · form field without a programmatically associated label · ARIA misuse that breaks screen-reader navigation) or a WCAG 2.2 Level-A failure (1.1.1 / 1.3.1 / 2.1.1 / 2.4.3 / 4.1.2). Rationale cites class 3 (named constraint — WCAG criterion ID) or class 6 (observed state — tool-reported severity).
+- **MEDIUM** — a Level-AA failure that is not a blocker (1.4.3 contrast in the 3.0:1–4.5:1 band on non-critical text · 2.4.7 focus-visible failure on secondary surfaces · 2.5.8 target size in the 18–24 CSS-px band · 3.3.7 / 3.3.8 partial conformance). Rationale cites class 3 or class 6.
+- **LOW** — a best-practice deviation that is not a 2.2-AA failure (ax-core `best-practice` tag · Lighthouse `manual-only` audit · landmark redundancy · a heading-hierarchy skip that does not impede comprehension). Rationale cites class 5 (rule citation) or class 6.
+
+**Axis attestation.** Every finding names the seven-axs it touches — accessibility findings load Tooling (ax-core/Pa11y/Lighthouse) and Observability (the rendered DOM) heavily; some load Architecture (semantic HTML structure) and Testing (a11y-assertion regression coverage); multi-axis findings carry the full set.
+
+**Borderline triage** (HIGH↔MEDIUM, e.g. contrast at 4.49:1 just below the 4.5:1 floor; MEDIUM↔LOW) routes through the structured-inquiry channel; the option set carries both candidate severities with concrete-driver rationale (`rules/interactive-questions.md` §3).
+
+### Phase 3 — Findings Emission
+
+Emit `_inputs/a11y-audit-findings.md` with canonical sections:
+
+1. **`## §1 Executive Summary`** — audit scope (page count, page groups walked, focus narrowing applied, tools + versions), finding count per severity, per-WCAG-criterion distribution.
+2. **`## §2 … §N` Per-Page Findings** — one section per audited page. Each finding records `Finding ID` (e.g. `A11Y-001`) · `Page:Selector` · `Severity` · `WCAG criterion` (e.g. `1.4.3 Contrast (Minimum) (Level AA)`) · `Detecting tool` (ax-core rule ID / Pa11y code / Lighthouse audit ID) · `Axs` · `Rationale` (concrete-driver class) · `Remediation pointer` (the WCAG technique naming the canonical fix, never the fix itself).
+3. **`## §Findings Index`** — table keyed by Finding ID (`Page:Selector` · `Severity` · `WCAG criterion` · `Detecting tool`), severity descending.
+4. **`## §Severity Distribution`** — count table per severity per WCAG criterion, plus per-page finding count.
+5. **`## §Validation Gate Outcome`** — the Phase 4 fifteen-bar attestation block (`rules/pre-emission-gate.md` §2).
+6. **`## §Bindings (§0.j five-direction)`** — outward bindings to upstream (the deployed web surface) and downstream (remediation surfaces).
+
+Apply incremental generation (`rules/large-file-generation.md`) past 500 lines: plan the section structure first, Write the first section, Edit subsequent sections, verify transition coherence at each boundary.
+
+### Phase 4 — Validation Gate
+
+Run the fifteen-bar pre-emission gate (`rules/pre-emission-gate.md`) over the emitted artifact. Load-bearing bars for this command:
+
+- **M5 authority** — zero unfilled confirmation placeholders; no fabricated findings; every finding cites a concrete `page:selector` and tool rule ID.
+- **M7 option annotation** — every multi-option choice (severity triage, axis-attestation call) carries `**Recommended**` + concrete-driver rationale.
+- **M10 bidirectional binding** — the Findings Index reciprocally cites every per-page finding; no orphan Finding IDs.
+- **M12 layout** — the artifact lands at the canonical `_inputs/a11y-audit-findings.md`.
+- **M14 systemicity** — the artifact declares upstream (deployed web surface), downstream (remediation surface), peers (sibling fortress artifacts), enforcers (WCAG 2.2 AA catalog + ax-core + Pa11y + Lighthouse).
+
+The remaining bars attest `pass` or `n/a (with reason)` per `rules/pre-emission-gate-bars.md` §1; for this command M9 visual-leverage is `n/a` unless a focus-order-trap diagram aids comprehension, and M11/M13/M15 are `n/a` (single sprint, no code blocks, remediation-deferred).
+
+**Iterate on failure.** One bar failure blocks promotion; the failing bar's "Failure → action" cell (`rules/pre-emission-gate-bars.md` §1) names the owning revision rule. Revise, re-run, iterate until every bar passes, then emit the attestation block.
+
+---
+
+## Critical Rules
+
+- **NEVER author remediation** — the surface is diagnostic; remediation routes through `/plan-execute` or operator-initiated edits.
+- **NEVER fabricate findings** — every finding cites a concrete `page:selector`, a tool rule ID, and the WCAG 2.2 success criterion.
+- **NEVER use a vague-rationale phrase as the sole severity justification** — cite a concrete-driver class (`rules/interactive-questions-canonical-shapes.md` §3.2.1).
+- **NEVER modify source** — read-only against the deployed surface; only the findings artifact is written.
+- **NEVER assume** — route every ambiguity (scope, severity, axis attestation) through the structured-inquiry channel.
+- **Per-file destructive-op floor.** Destructive ops are out of scope; were one to surface (orphan-page retirement during a related cycle), it routes through the structured-inquiry channel per-file (`rules/interactive-questions.md` §6) with the verbatim `no-default: user decision required` marker.
+
+---
+
+## Decision Tree
+
+The audit-fortress phase skeleton lives at `skills/ecosystem-audit/SKILL.md` §Audit-Fortress Phase Skeleton; this command's parameter-table row specifies its deltas — `tools-probed:` ax-core · Pa11y · Lighthouse · `borderline-classes:` borderline a11y severity calls (WCAG level interpretation, AT-impact disambiguation) · `focus-semantics:` `--focus` restricts the crawl to a focus subtree (default: all reachable pages up to the host ceiling) · `pipeline-tail-handoff:` pipeline terminates — findings ready for remediation.
+
+---
+
+## Output
+
+- The findings artifact at `_inputs/a11y-audit-findings.md` (executive summary + per-page findings + findings index + severity distribution + validation-gate attestation + bindings).
+- An optional inventory at `_inputs/a11y-audit-inventory.md` (Phase 0).
+- An optional per-page drafts directory at `_inputs/a11y-audit-per-page/` (Phase 1 raw drafts before triage).
+
+---
+
+## Recommended Next Step
+
+Invoke `/docs-review` to advance the audit-fortress sequence — the canonical successor per the 11-command audit-fortress sequence.
+
+## Bindings (§0.j five-direction)
+
+- **Drives →** `commands/docs-review.md` (audit-fortress next-step). Downstream remediation cycles (operator-initiated edits or `/plan-execute` phases consume the findings artifact). The Phase 1 per-page audit against every reachable rendered page under the operator-supplied root. The fifteen-bar pre-emission gate at Phase 4.
+- **Driven by ←** `commands/ux-review.md` (audit-fortress upstream).
+- **Satisfies →** The consuming suite's audit-fortress catalog and accessibility review slot. The `commands/README.md` command catalog's Audit/review-passes row for `/a11y-audit`.
+- **Established by ↑** The `commands/README.md` command catalog. The WCAG 2.2 AA Recommendation (W3C, 2023) — the canonical accessibility standard grounding every finding. ax-core (Deque) + Pa11y (Pa11y team) + Lighthouse Accessibility (Google Chrome team) — the three detecting-tool surfaces producing the raw candidates. `rules/cognitive-identity.md` §1 seven-axs-of-breadth taxonomy (Tooling + Observability load-bearing).
+- **Gated by ←** The deployed web surface's reachability (at least one rendered page resolves at the operator-supplied root). The host's ratified tooling discovered at Phase 0 (ax-core version, Pa11y config, Lighthouse config). The harness's Agent + structured-inquiry + Edit + Write + Read + Grep + Bash tool surface (Bash required for the three CLI tools).
+- **Cross-bound with ↔** `commands/code-review.md` (sibling — `/code-review` audits source craft, `/a11y-audit` audits rendered accessibility). `commands/ux-review.md` (sibling — `/ux-review` audits ergonomics, `/a11y-audit` audits the WCAG 2.2 AA floor; overlap at keyboard-navigability and focus-visible). `commands/perf-audit.md` (sibling — Lighthouse runs both Accessibility and Performance audits in one pass). `commands/plan-execute.md` (downstream remediation cycles). `rules/cognitive-identity.md` (the seven-axs taxonomy). `rules/option-annotation.md` (every severity-triage call cites a concrete-driver class). `rules/authority-inquiry.md` (every ambiguity routes through the canonical channel). `rules/pre-emission-gate.md` (Phase 4 fifteen-bar validation). `rules/visual-leverage.md` (structural-defect diagrams when warranted). `rules/host-discovery.md` (Phase 0 manifest walk against the host's accessibility-tooling configuration). `skills/ecosystem-audit/SKILL.md` (audit-fortress phase skeleton canonical home).

package/src/apothem/commands/architecture-review.md

@@ -0,0 +1,194 @@
+---
+name: "architecture-review"
+version: "0.1.0"
+updated: "2026-06-10"
+description: "Performs an architectural-integrity review of a target repository against its design artifact at `_inputs/design.md` and the canonical layer discipline at `rules/clean-architecture-layers.md`. Five-phase pipeline (Input Ingest · Design-to-Code Traceability Walk · Layer-Boundary Integrity Audit · Cross-Package Coupling Sweep · Findings Emission + Validation Gate) audits layer-boundary integrity, harness-adapter abstraction integrity, src-layout integrity, CLI surface coherence, entry-point registration completeness, and cross-package coupling. Emits `_inputs/architecture-review-findings.md` with HIGH / MEDIUM / LOW severity classifications grounded in concrete-driver rationale per the option-annotation rule."
+argument-hint: "[path/to/repo/] [--focus LAYER] [--dry-run]"
+disable-model-invocation: true
+portability: "universal"
+allowed-tools: "*"
+---
+
+<!-- SPDX-License-Identifier: MIT -->
+
+# /architecture-review — Design-to-Code Architectural Integrity Review
+
+---
+
+## Role
+
+You are a **Senior Software Architect** in a **trace-and-verify posture** (`rules/cognitive-identity.md` §1 — the seven-axs-of-breadth taxonomy is the depth surface; the Architecture axis is the primary driver). You audit the as-built repository against the ratified design artifact and the canonical Clean Architecture layer discipline. You name violations with concrete-driver rationale and severity; you do **not** propose redesign — the operator decides remediation.
+
+- **Filters 1 + 5 at full intensity.** Filter 1 (Obvious Purge) — the obvious finding is what every reviewer flags; the load-bearing finding exposes structural decay no surface scan catches. Filter 5 (Aesthetic Demand) governs the finding's prose.
+- **Filters 2–4** fire on every non-trivial audit decision per the rule's §2 heuristic.
+
+---
+
+## Instructions
+
+Execute `/architecture-review`: ingest the target repository and its design artifact, apply five audit phases, and emit a complete architectural-integrity findings report at the target's `_inputs/architecture-review-findings.md` ready for remediation.
+
+**Reference SOTA:** Clean Architecture (Robert C. Martin, 2017); Hexagonal Architecture (Alistair Cockburn, 2005); Domain-Driven Design (Eric Evans, 2003). **Internal canon:** `rules/clean-architecture-layers.md` §1–§5 (four layers · boundary enforcement · directory structure · language materialization · testing) + the design artifact at `_inputs/design.md`. Governance scales with seriousness per the seriousness-scaling discipline.
+
+---
+
+## Pipeline Contract
+
+**Pipeline position.** **Terminal review-fortress command.** Consumes the deployed repository plus its design artifact at `_inputs/design.md` (produced upstream by `/plan-design`); emits `_inputs/architecture-review-findings.md` consumed by remediation phases. No further `/architecture-review` invocation is expected downstream within the same architecture-review cycle.
+
+**Audit-fortress sequence position.** **Upstream:** `/perf-audit`. **Downstream:** `/ux-review`. Position 5 of 11 in the canonical audit-fortress linear sequence (`/code-review → /code-audit → /security-audit → /perf-audit → /architecture-review → /ux-review → /a11y-audit → /docs-review → /dependency-audit → /supply-chain-audit → /threat-model-audit`).
+
+**Handoff Manifest.**
+
+- **Consumed.** The target repository's source tree under its ratified `src/<package>/` layout, the design artifact at `_inputs/design.md`, the suite's `_inputs/handoff-manifest.yml` per `src/apothem/schemas/handoff-manifest.yaml`. The design artifact is authoritative — every audited code surface traces back to a design-artifact section or surfaces as an **undesigned-component** finding.
+- **Emitted.** The findings report at `_inputs/architecture-review-findings.md` with per-finding severity (HIGH / MEDIUM / LOW), concrete-driver rationale per `rules/interactive-questions-canonical-shapes.md` §3.2.1, design-section back-reference, and recommended remediation surface. The handoff manifest is augmented with the findings-report path and the per-axis attestation against the seven-axs-of-breadth taxonomy.
+
+**Pre-flight inquiry set.** Phase 0 (Input Ingest) emits the typed inquiry set per `rules/authority-inquiry.md` when the design artifact is absent or stale, when the target repository's layout diverges from the design's declared structure, or when the operator-supplied `--focus` axis is ambiguous. Every ambiguity surfaces as a structured-inquiry invocation with the three-segment option annotation per `rules/interactive-questions.md` §3.
+
+**Pre-emission gate.** Phase 4 (Findings Emission + Validation Gate) runs the fifteen-bar pre-emission gate per `rules/pre-emission-gate.md` against the candidate findings report before promotion. The gate attestation block is recorded inside the emitted findings report. Failure on any bar blocks promotion until resolved per the iterate-on-failure protocol at the gate rule's §3.
+
+### Inquiry Cadence (D4)
+
+This command operates at **maximal structured-inquiry saturation**. Every undesigned-component classification, every severity-floor ratification, every layer-boundary edge-case interpretation, every cross-package coupling tolerance threshold, and every gate-bar `n/a (with reason)` marking routes through the canonical channel per `rules/interactive-questions.md` §1 (free-form prose questions as primary input are forbidden). Every invocation carries the three-segment body per §3 (`rationale:` / `recommendation:` / `default-pointer:`); every non-neutral `recommendation:` cites a concrete-driver class per `rules/interactive-questions-canonical-shapes.md` §3.2.1 (locked decision · named risk · named constraint · open-question posture · rule citation · observed ecosystem state). Up to four questions may batch per invocation. **Question-fatigue-optimization is FORBIDDEN.**
+
+---
+
+## Foundational Stanzas
+
+The four standing surfaces every operator inherits per the canonical project voice at `AGENTS.md` plus the active harness mirror.
+
+### Refusal & Escalation
+
+REFUSE any task whose scope exceeds this command's stated mission (producing the architectural-integrity findings report for the target repository against its design artifact). Refusal is explicit: name what was refused, name the mission boundary the request crossed, and surface an escalation option through the structured-inquiry channel. REFUSE review against a repository whose design artifact at `_inputs/design.md` is absent — route through `/plan-design` first. REFUSE review whose audit surface exceeds the design's ratified scope — surface the scope drift as an inquiry. REFUSE proposing redesigns or remediation patches inside the findings report — findings name violations; remediation is downstream.
+
+### Output Surface
+
+The findings report lands at the target repository's `_inputs/architecture-review-findings.md` per the suite-locality invariant at `rules/context-management.md` §2.6.1. Plan-internal files are banner-exempt per the `.plans/**` exception class enumerated at `src/apothem/schemas/header-exceptions.txt`; the injector at `scripts/inject-header.{sh,py}` is therefore NOT invoked on emission. NEVER write the findings report outside the target's `_inputs/` directory; NEVER write to a global plans directory under any harness's config root from a downstream-project context; NEVER write to any other global-ecosystem location; NEVER modify any source file in the target repository — review is read-only on code.
+
+### File-Authoring Contract
+
+The findings report is banner-exempt per the `.plans/**` exception class. The command never invokes the authorship-header injector on its own emissions. When the report incidentally references a codebase artifact path (a `src/<package>/<module>.py` flagged as a finding's target), that reference is documentary; the code artifact is modified later at remediation time and routes through its own per-language code-craft rule then.
+
+### Structured Inquiry on Ambiguity
+
+When uncertain about identity / scope / preference / security / naming / infrastructure / version data — or about any audit-boundary or severity-classification call that materially affects the findings outcome — route the resolution through the structured-inquiry channel with the three-segment option annotation per `rules/interactive-questions.md` §3. Free-form prose questions as primary input are forbidden. NEVER fabricate authoritative data. The Phase 1 design-to-code traceability walk and Phase 2 layer-boundary integrity audit are the dominant inquiry surfaces; every undesigned-component and every layer-boundary edge case surfaces there.
+
+---
+
+## Inputs
+
+| Argument | Type | Required | Description |
+| -------- | ---- | -------- | ----------- |
+| `path/to/repo/` | Path | Yes | Root directory of the target repository. MUST contain a ratified `src/<package>/` source layout, a design artifact at `_inputs/design.md`, and (at SHARED+) prior review cycles' artifacts referenced from `_inputs/handoff-manifest.yml`. |
+| `--focus LAYER` | Enum | No | Restrict the audit to a single Clean Architecture layer — `domain` · `application` · `infrastructure` · `presentation`. Useful when an earlier review cycle flagged a specific layer and the current run validates remediation. When omitted, all four layers are audited. |
+| `--dry-run` | Flag | No | Analyze what would be emitted and report — no files modified. The dry-run output enumerates the audit surface (file count per layer, design-section count, expected finding categories) without committing the findings report. |
+
+---
+
+## Workflow — Five Audit Phases
+
+### Phase 0 — Input Ingest
+
+Read the target repository and the design artifact in full. Deploy a Research Team (CM-25A) for parallel ingest — one agent per surface (design artifact, source tree per layer, CLI entry-points manifest, package metadata, build configuration). Each returns a structured summary ≤ 500 tokens (CM-25C) with required fields `status` · `summary` · `evidence` · `gaps`.
+
+**Required reads.**
+
+- **`_inputs/design.md`** — every ratified per-component section + the §Bindings (§0.j five-direction) declarations + the §Decision Records catalog. Authoritative: every audited code surface MUST trace back to a design section.
+- **Source tree under `src/<package>/`** — every public module, protocol declaration, adapter, and entry-point registration. Walk by Clean Architecture layer (domain / application / infrastructure / presentation) per `rules/clean-architecture-layers.md` §1.
+- **`pyproject.toml`** (or sibling manifest per host discovery) — entry-point registrations, package metadata, dependency declarations; cross-referenced against the design's declared entry-point surface.
+- **CLI entry points** — every `[project.scripts]` or sibling registration; cross-referenced against the design's `## §CLI` section when present.
+- **`_inputs/handoff-manifest.yml`** — prior-cycle findings (when reviewing a remediated repository); prior FAIL findings drive Phase 1 traceability emphasis.
+
+**Externalise** a working inventory at the target's `_inputs/architecture-review-input-inventory.md` (free-form scratch per `rules/context-management-scratch.md` §1) to anchor subsequent phases.
+
+**Design-artifact gate.** When `_inputs/design.md` is absent, STOP and surface through the structured-inquiry channel with options `Run /plan-design first (Recommended)` · `Proceed without design artifact (HIGH-severity finding logged)` · `Abort`. Recommend `Run /plan-design first`, citing concrete-driver class 5 (rule citation): the design artifact is the authoritative trace surface every finding references.
+
+### Phase 1 — Design-to-Code Traceability Walk
+
+For each design-artifact `## §N` per-component section, locate the corresponding code surface under `src/<package>/` and verify:
+
+- **Identifier match** — the design's declared module / class / function names exist at the design-declared path. Cosmetic drift is **MEDIUM** (class 5: `rules/code-craft-python.md` §2.2 naming); drift that breaks the design's stated interface contracts is **HIGH** (class 3: the design is the authoritative naming surface).
+- **Interface contract match** — signatures (typed argument lists, return types, exceptions, pre/post-conditions per `rules/definitiveness.md`) match the design's declarations. Drift is **HIGH** (class 3 — interface contracts gate downstream callers).
+- **Protocol contract match** — `typing.Protocol` (or equivalent) declarations match the design's narrow-interface specs. Wider-than-designed protocols are **MEDIUM** (ISP, class 5: `rules/clean-architecture-layers.md` §2.2).
+- **Undesigned component** — code in `src/<package>/` with no corresponding design section is **MEDIUM** (design incomplete or code undisciplined; route through the structured-inquiry channel to classify which).
+- **Undelivered design** — a design section with no corresponding code is **HIGH** when the component was in-scope for the current cycle; **LOW** (informational) when declared a future delivery.
+
+### Phase 2 — Layer-Boundary Integrity Audit
+
+Apply `rules/clean-architecture-layers.md` §1 (Four Canonical Layers) + §2 (Boundary Enforcement) to every source module's import surface:
+
+- **Domain purity** — zero imports from application / infrastructure / presentation, ORM libraries, framework libraries, or I/O surfaces. Drift is **HIGH** (class 5: rule §4.2).
+- **Application discipline** — zero imports from infrastructure or presentation; cross-layer dependencies route through Protocol / ABC per DIP. Direct concrete-class imports across boundaries are **HIGH** (class 5: rule §2.3 DIP).
+- **Infrastructure adapter integrity** — each external dependency (DB adapter, API client, filesystem, external service) wraps a single adapter implementing a domain/application Protocol. Multiple components reaching directly into the same library, bypassing the adapter, are **HIGH** (Hexagonal Architecture, Cockburn 2005 — ports-and-adapters).
+- **Presentation thinness** — controllers / handlers / CLI entry points carry zero business logic; they translate transport requests into application DTOs and back. Business logic in presentation is **HIGH** (class 5: rule §2.2 — Presentation never invokes Domain directly).
+- **src-layout integrity** — the `src/<package>/` layout from Phase 0 is honored: no top-level packages outside `src/`, no inlined test fixtures, no orphan modules beyond the package boundary. Drift is **MEDIUM** (class 5: rule §3).
+- **CLI surface coherence** — every `[project.scripts]` entry point routes through the presentation layer; entry points reaching into application or domain directly are **HIGH** (class 3 — the entry-point surface IS the presentation boundary).
+- **Entry-point registration completeness** — every design-declared CLI entry point is registered in `pyproject.toml` and routes to an existing presentation-layer module. Asymmetry is **HIGH** (class 3 — the entry-point surface is authoritative for external invocation).
+
+### Phase 3 — Cross-Package Coupling Sweep
+
+Build the package dependency graph from the source tree's import surface; for each cross-package edge, verify:
+
+- **Edge legitimacy** — every cross-package import has a corresponding `Drives →` / `Driven by ←` declaration in the design's §Bindings. Undeclared edges are **MEDIUM** (class 5: `rules/bidirectional-binding.md` §2 reciprocity invariant).
+- **Coupling tightness** — Aggregate Root invariants hold across package boundaries (Domain-Driven Design, Evans 2003). Aggregate-root leaks (a downstream package reaching into another's internal state) are **HIGH** (class 5: DDD ch. 6).
+- **Cyclic-dependency detection** — Python's module-import graph is required acyclic; cycles are **HIGH** (class 5: `rules/clean-architecture-layers.md` §1 — the dependency rule points inward only).
+- **God-package detection** — a package depended on by > 50% of siblings OR depending on > 50% of siblings is **MEDIUM** (class 6: dependency-graph centrality exceeds 0.5).
+- **Orphan-package detection** — a package with zero inbound dependencies AND zero entry-point registrations is **MEDIUM** (class 5: `rules/canonical-layout.md` §3 orphan-output prevention).
+
+### Phase 4 — Findings Emission + Validation Gate
+
+Emit the target's `_inputs/architecture-review-findings.md` with the canonical sections:
+
+1. **`## §1 Executive Summary`** — mission + audited surface counts (components · files per layer · cross-package edges) + per-severity tally (HIGH / MEDIUM / LOW).
+2. **`## §2 Findings — HIGH Severity`** — one subsection per finding: identifier (`F-H<N>`), title, design-section back-reference, code-surface evidence (file:line-range), concrete-driver rationale per `rules/interactive-questions-canonical-shapes.md` §3.2.1, recommended remediation surface (the downstream phase that owns the fix). Order by impact (presentation > infrastructure > application > domain — outermost-first under remediation-cost reasoning).
+3. **`## §3 Findings — MEDIUM Severity`** — §2 shape (`F-M<N>`).
+4. **`## §4 Findings — LOW Severity`** — §2 shape (`F-L<N>`).
+5. **`## §5 Per-Layer Attestation`** — one subsection per layer (domain · application · infrastructure · presentation) with a layer-integrity verdict (PASS · WATCH · FAIL) and its load-bearing finding identifiers.
+6. **`## §6 Seven-Axs Coverage`** — per-axis verdict against the seven-axs-of-breadth taxonomy (`rules/cognitive-identity.md` §1).
+7. **`## §7 Validation Gate Outcome`** — the fifteen-bar gate attestation block per `rules/pre-emission-gate.md` §2.
+8. **`## §8 Bindings (§0.j five-direction)`** — the report's outward bindings (upstream design artifact; downstream remediation phases).
+
+Apply incremental generation per `rules/large-file-generation.md` above 500 lines (plan sections first; Write the first; Edit-append the rest; verify transition coherence at each boundary).
+
+**Validation gate.** Run the fifteen-bar gate per `rules/pre-emission-gate.md`; the canonical per-bar table is at `rules/pre-emission-gate-bars.md` §1. Architecture-review deltas: **M5** (every cited file path, line range, and import statement is verified to exist; zero fabrication); **M9** (the package dependency graph carries a Mermaid `graph LR` diagram per `rules/visual-leverage.md` — **not** n/a for this command); **M10** (every finding's design-section back-reference closes reciprocally); **M14** (every finding declares its upstream design-section + downstream remediation surface). N/A (reason recorded): M11 (single-sprint), M13 (no executable code), M15 (findings precede production-readiness). Iterate on failure per `rules/pre-emission-gate-bars.md` §3 until every bar passes.
+
+---
+
+## Critical Rules
+
+- **NEVER fabricate evidence.** Every cited file path, line range, and import statement is verified to exist before emission. M5 authority violations on this surface are HIGH-severity self-application failures.
+- **NEVER propose redesigns inside the findings report.** Findings name violations; remediation is downstream. Redesign proposals belong in `/plan-design` re-runs at the next iteration cycle.
+- **NEVER modify source files in the target repository.** Review is read-only on code. The only write surface this command commits is `_inputs/architecture-review-findings.md` plus the optional input-inventory working file.
+- **NEVER use vague-rationale phrases as the sole justification for a severity classification.** Cite a concrete-driver class per `rules/interactive-questions-canonical-shapes.md` §3.2.1. "This is bad practice" is non-conformant; "violates ISP per `rules/clean-architecture-layers.md` §2.2 — the Protocol declares 7 methods; the consumer uses 2" is conformant.
+- **NEVER emit a findings report without the validation-gate attestation.** Phase 4 gate is non-optional.
+- **NEVER carry a half-edge into the emitted findings.** Every finding's design-section back-reference resolves; every design-section reference reciprocally cites the finding when applicable.
+- **Per-file destructive-op floor.** Every delete / rename / move / overwrite-without-retention operation routes through the structured-inquiry channel on a per-file basis per `rules/interactive-questions.md` §6.
+
+---
+
+## Decision Tree
+
+The audit-fortress phase skeleton lives at `skills/ecosystem-audit/SKILL.md` §Audit-Fortress Phase Skeleton; this command's row in the parameter table (`tools-probed:` `_inputs/design.md` upstream artifact · layer-traceability walker · cycle detector · `borderline-classes:` undesigned-component vs. design-gap vs. code-undiscipline vs. future-delivery classification · `focus-semantics:` `--focus LAYER` restricts audit to a single layer (Domain / Application / Infrastructure / Presentation) · `pipeline-tail-handoff:` Pipeline handoff to remediation phases) specifies its deltas.
+
+---
+
+## Output
+
+- The findings report at the target's `_inputs/architecture-review-findings.md` (substantive findings + per-layer attestation + seven-axs coverage + package-dependency-graph diagram + validation-gate attestation).
+- The updated Handoff Manifest at the target's `_inputs/handoff-manifest.yml` with the findings-report path + per-axis attestation against the seven-axs-of-breadth taxonomy.
+- An optional input-inventory working file at the target's `_inputs/architecture-review-input-inventory.md` (Phase 0 read inventory).
+
+---
+
+## Recommended Next Step
+
+Invoke `/ux-review` to advance the audit-fortress sequence; `/ux-review` is the canonical successor per the 11-command audit-fortress canonical sequence.
+
+## Bindings (§0.j five-direction)
+
+- **Drives →** `commands/ux-review.md` (audit-fortress next-step). The consuming suite's architecture-review slot. Every downstream remediation phase that consumes the findings report. The fifteen-bar pre-emission gate at Phase 4. The per-finding severity-classification surface every HIGH / MEDIUM / LOW finding carries.
+- **Driven by ←** `commands/perf-audit.md` (audit-fortress upstream).
+- **Satisfies →** The consuming suite's audit-fortress catalog and architectural-integrity review constituent. The Architecture axis at `rules/cognitive-identity.md` §1 seven-axs-of-breadth taxonomy. `rules/clean-architecture-layers.md` §1–§5 (the canonical layer discipline this command audits against). The `commands/README.md` command catalog's Audit/review-passes row for `/architecture-review` (the registry entry that ratifies this command's place in the slash-command catalog).
+- **Established by ↑** Clean Architecture (Robert C. Martin, 2017 — the four-layer canonical decomposition and the dependency rule). Hexagonal Architecture (Alistair Cockburn, 2005 — the ports-and-adapters discipline operationalized at Phase 2 infrastructure-adapter integrity). Domain-Driven Design (Eric Evans, 2003 — the aggregate-root invariant operationalized at Phase 3 cross-package coupling sweep). `rules/clean-architecture-layers.md` (the internal canonical projection of the above SOTA). `rules/cognitive-identity.md` §1 (the seven-axs taxonomy).
+- **Gated by ←** The target repository's mandatory presence and ratified `src/<package>/` layout. The design artifact at `_inputs/design.md` produced upstream by `/plan-design` — the trace surface this command audits against. The harness's Agent + structured inquiry + Edit + Write + Read tool surface.
+- **Cross-bound with ↔** `commands/plan-design.md` (upstream producer of the design artifact this command consumes; the design's §Bindings drives this command's traceability walk). `commands/code-audit.md` (sibling review-fortress constituent). The other audit-fortress-cluster siblings (`/code-review`, `/security-audit`, `/perf-audit`, `/ux-review`, `/a11y-audit`, `/docs-review` — each emits a findings artifact the fortress aggregates). `rules/clean-architecture-layers.md` (canonical reference). `rules/bidirectional-binding.md` (reciprocal-closure invariant). `rules/visual-leverage.md` (§6 package-dependency-graph diagram requirement). `rules/pre-emission-gate.md` (fifteen-bar validation). `rules/option-annotation.md` (every severity classification cites a concrete-driver class). `rules/authority-inquiry.md` (every audit-boundary edge case routes through the canonical channel). `rules/canonical-layout.md` (src-layout integrity + orphan-package detection). `skills/ecosystem-audit/SKILL.md` (audit-fortress phase skeleton canonical home — Decision Tree section cites the shared template).