Merge phase/05-atelier-mcp — v1.17.5 (v1.18 P5 Atelier MCP server complete)

REQ-223: mcp/atelier/server.py plugin-registry MCP server (stdio, D-135).
NovaAtelierServer wraps MCPServer (SDK v2, D-137) if installed; degrades
to _ToolRegistry fallback if SDK absent (testable in CI without SDK).
plugins/principles.py (lookup_principle, list_domains, matrix_lookup) +
plugins/validation.py (validate_against_principles — agentic validation
beyond Wiz/Checkmarx/Mend). 4 tools, 2 plugins.

REQ-224: mcp/atelier/vendor/ pinned Atelier v0.3.6 (D-136) — core/
first-principles, domains/security/first-principles, review/agent-checklist,
matrix/principles-matrix. vendor/VERSION.md + scripts/update_atelier_vendor.sh
for intentional upgrades. mcp/atelier/README.md (tools, architecture,
running, vendoring, extensibility, transport).

REQ-225: tests/test_atelier_mcp.py — 16 tests, all pass. Covers: plugin
discovery (both loaded), 4 tools registered, lookup_security_P4 (+P1,
unknown domain/principle), list_domains (19, security-relevant, ui-ux-not),
matrix_lookup (security 10 P-rules, unknown), validation (good-passes,
bad-secret-fails, bad-swallowed-error-fails, bad-obfuscated-names-fails,
result-structure).

---ci---
project: acdl
phase: 5
milestone: v1.18
status: execute
requirements:
  covered: [REQ-223, REQ-224, REQ-225]
  partial: []
---/ci---

.ciagent/PROJECT.md

@@ -980,7 +980,7 @@ or user-directed scope). New v1.7 decisions:
 | W1.A | AI-refinement trigger | **Accept recommendation.** Joint condition: N ≥ 50 consecutive changes with zero rollbacks AND no L1/L2 incident in last 6 months AND Infra & Ops unilateral override. |
 | W1.B | Multi-stack edge case rule | **Accept recommendation.** Permitted only for (a) DR-region mirror, (b) time-boxed experimental stack with TTL ≤ 30d, (c) explicit Infra & Ops approval with `multiStack.justification`. |
 | W2.A | Tag mutability for prod | **Accept recommendation (Path B).** Tag for dev/qa, SHA for prod. Platform CLI resolves tag→SHA for prod-bound workflows. Justified by the "Audit truth lives outside the repository" bet. |
-| BA.A | Initial L3B skill catalog | **Accept recommendation.** 5 skills: web API, worker, scheduled job, static asset, basic observability bootstrap. Addition criteria: (a) reviewable for sensitive data, (b) expressible as a single contract submission, (c) documented use case. |
+| BA.A | Initial L3B skill catalog | **Accept recommendation.** 5 skills: web API, worker, scheduled job, static asset, basic observability bootstrap. Addition criteria: (a) reviewable for sensitive data, (b) expressible as a single contract submission, (c) documented use case. **Extended v1.18 (REQ-221/222):** the BA.A 5-skill catalog is extended with 9 Atelier-derived production-grade engineering skills under `skills/` (api, security, data, testing, observability, errors, devops, infrastructure-as-code, compliance), indexed by `docs/skills.md`. The Atelier skills extend, not replace, the BA.A catalog. |
 | W3.D | L1/L2 standard versioning | **Decided.** Semver: interface → MAJOR, behavior → MINOR, lifecycle → PATCH (same as the v1.0 demo D-rule, lifted to the real platform). Pin model: L2 contracts pin L1 by `name@semver`; the resolver picks the highest compatible. Evolution: MAJOR bumps require a new registry entry (immutable publication); old entry enters a 12-month deprecation window. |
 | W3.E | Schema mandatory vs optional inputs | **Decided.** Per-env mandatory table: dev requires `stack` + `environment`; qa adds `validation.e2eSuite` + `validation.loadTest`; prod adds `runbook` + `dashboard` + `oncall`; dr adds `drDrillRef`. `inputs` map is always optional. `profile: agentic` fields (`naturalLanguageIntent`, `confidenceAtSubmission`, `agentTrace`) optional everywhere. |
 | BA.B | Confidence threshold tuning | **Decided.** Starting thresholds frozen for v1. Tuning begins in v1.2: track FP/FN per environment quarterly; override authority = Infra & Ops + SRE joint sign-off; any override is itself a confidence-event in the audit stream. |

docs/skills.md

@@ -0,0 +1,88 @@
+# Skills — Production-Grade Guidance for the Citizen Developer
+
+> **Source of truth (v1.18, REQ-221, REQ-222).** The Nova skill catalog
+> extends the BA.A 5-skill catalog (web API, worker, scheduled job, static
+> asset, basic observability bootstrap) with Atelier-derived production-
+> grade engineering principles. Each skill is a markdown file under
+> `skills/` keyed to an Atelier domain path.
+
+## How the Citizen Developer's AI Agent Consumes Skills
+
+1. **Before completing a task**, read the relevant skill file(s) that
+   match the task's domain.
+2. **Run `review/agent-checklist.md`** (from Atelier) before finishing —
+   the checklist items are the gate between "the code is written" and
+   "the task is done."
+3. **Use the Atelier MCP server** (`mcp/atelier/server.py`, P5) for
+   agentic validation — the `atelier.validate_against_principles` tool
+   catches correctness/clarity/simplicity/observability gaps that
+   deterministic scanners (Wiz, Checkmarx, Mend) cannot.
+
+## The 9 Skills
+
+| Skill | Atelier Source | Core Principles | BA.A Mapping |
+|---|---|---|---|
+| [`api.md`](../skills/api.md) | `domains/api/` | C1, C2, C6 | web API |
+| [`security.md`](../skills/security.md) | `domains/security/` | C1 | cross-cutting (all 5) |
+| [`data.md`](../skills/data.md) | `domains/data/` | C1, C4, C6 | web API, worker, scheduled job |
+| [`testing.md`](../skills/testing.md) | `domains/testing/` | C1, C5 | UAT (citizen-dev RACI) |
+| [`observability.md`](../skills/observability.md) | `domains/observability/` | C7 | basic observability bootstrap |
+| [`errors.md`](../skills/errors.md) | `domains/errors/` | C1, C7 | web API, worker, scheduled job |
+| [`devops.md`](../skills/devops.md) | `domains/devops/` | C5, C7, C8 | scheduled job, worker |
+| [`infrastructure-as-code.md`](../skills/infrastructure-as-code.md) | `domains/infrastructure-as-code/` | C1, C5, C8 | static asset |
+| [`compliance.md`](../skills/compliance.md) | `domains/compliance/` | C1, C5 | cross-cutting (all 5) |
+
+## Atelier Provenance
+
+The skills are derived from [Atelier](https://git.cloudinit.dev/coreci/atelier)
+— a first-principles docs-as-code engineering framework with 8 core
+principles (C1–C8) and 19 domains, each with 10 derived P-rules. The
+skills distill the citizen-developer-relevant subset of each domain's
+first-principles, link to the agent-checklist triggers, and map to the
+existing BA.A catalog.
+
+Atelier is vendored under `mcp/atelier/vendor/` (pinned tag, D-136) for
+audit reproducibility — an agentic validation result is replayable
+against the exact principles that produced it.
+
+## The 8 Core Principles (from Atelier)
+
+| # | Principle | One-line |
+|---|---|---|
+| C1 | Correctness | The system does what it is supposed to do, and nothing else. |
+| C2 | Clarity | The intent of the code is obvious to its reader. |
+| C3 | Simplicity | The solution is as simple as possible, and no simpler. |
+| C4 | Locality | Decisions and their consequences live near each other. |
+| C5 | Reversibility | Every decision can be undone, and the cost of undoing is known. |
+| C6 | Composability | Parts combine into wholes, and the parts are reusable. |
+| C7 | Observability | The system's behavior is visible to those who must understand it. |
+| C8 | Economy | The system uses no more resources than the task requires. |
+
+Precedence: C1 > C2 > C3 > C4 > C5 > C6 > C7 > C8. Correctness is never
+sacrificed.
+
+## Reference-Only Domains (cited inside skills, not elevated to skill files)
+
+These 4 Atelier domains are relevant to a citizen developer but are cited
+inside the 9 skills above rather than getting their own skill file:
+
+- **Performance** (`domains/performance/`) — cited in `observability.md` +
+  `devops.md` (bounded operations, timeouts, N+1)
+- **Documentation** (`domains/documentation/`) — the runbook requirement
+  (W3.E prod mandatory) is the documentation skill in practice
+- **Concurrency** (`domains/concurrency/`) — cited in `errors.md` +
+  `devops.md` (bounded queues, cancellation, timeout)
+- **AI/ML** (`domains/ai-ml/`) — scope: engineering discipline (data
+  versioning, evaluation, serving, drift), not algorithm design
+
+## Excluded Domains (not relevant to Nova citizen developer)
+
+6 Atelier domains are excluded from the Nova skill catalog (not relevant
+to a citizen developer building on Nova's infrastructure platform):
+
+- UI/UX — Nova has no frontend (frontend-engineer deactivated, PERSONAS.md)
+- Kubernetes — Nova is AWS-only this milestone (NORTH_STAR Non-Goal #7)
+- GitOps + Operators — future roadmap (no GitOps reconciler today)
+- Edge — not in scope (Nova is cloud, not edge)
+- Messaging — not in scope (Nova deploys infra, not message brokers)
+- i18n — application-level concern, not infrastructure

mcp/__init__.py

@@ -0,0 +1 @@
+# mcp/atelier — Nova Atelier MCP server package (v1.18)

mcp/atelier/README.md

@@ -0,0 +1,96 @@
+# Nova Atelier MCP Server
+
+> **v1.18, REQ-223, REQ-224.** An MCP (Model Context Protocol) server that
+> exposes Atelier engineering principles to the citizen developer's AI
+> agent. Plugin-registry architecture (D-140); stdio transport (D-135);
+> vendored Atelier (D-136) for audit reproducibility.
+
+## What This Is
+
+The server exposes 4 tools that let a citizen developer's AI coding agent
+look up production-grade engineering principles and validate code against
+them — agentic validation that goes **beyond deterministic scanners**
+(Wiz, Checkmarx, Mend) by catching correctness, clarity, simplicity, and
+observability gaps.
+
+## Tools
+
+| Tool | Description |
+|---|---|
+| `atelier.lookup_principle(domain, principle_id)` | Look up a principle by domain + P-rule ID (e.g., `security`, `P4`). Returns the principle text + the core C-rule it derives from. |
+| `atelier.list_domains()` | List the 19 Atelier domains with P-rule counts + Nova-relevance. |
+| `atelier.matrix_lookup(domain)` | Look up the domain→core principle mapping for a given domain. |
+| `atelier.validate_against_principles(snippet, domains?)` | Validate a code/diff snippet against the Atelier agent-checklist. Returns pass/fail per check item with the principle citation. |
+
+## Architecture — Plugin Registry (D-140)
+
+```
+mcp/atelier/
+├── server.py              # entrypoint: loads plugins, starts server
+├── plugins/
+│   ├── __init__.py
+│   ├── principles.py       # lookup_principle, list_domains, matrix_lookup
+│   └── validation.py       # validate_against_principles
+├── vendor/                 # pinned Atelier snapshot (D-136)
+│   ├── VERSION.md           # pinned tag + upgrade instructions
+│   ├── core/first-principles.md
+│   ├── domains/security/first-principles.md
+│   ├── review/agent-checklist.md
+│   └── matrix/principles-matrix.md
+└── README.md               # this file
+```
+
+Each plugin module exposes `register(mcp) -> None` and calls `@mcp.tool()`
+for its tools. `server.py` scans `plugins/` and calls `register` on each.
+**Future capabilities drop in as a new plugin file — no `server.py` edits.**
+
+## Running
+
+### With the MCP Python SDK installed
+
+```bash
+pip install "mcp[cli]"
+python3 -m mcp.atelier.server
+```
+
+The server runs over stdio. An MCP client (e.g., the citizen developer's
+AI coding agent) spawns it as a subprocess and calls tools via JSON-RPC.
+
+### Without the SDK (fallback / test mode)
+
+The server degrades to a plain-Python tool registry. Tools are callable
+directly — this is how tests run without the SDK installed:
+
+```python
+from mcp.atelier.server import NovaAtelierServer
+s = NovaAtelierServer()
+s.load_plugins()
+result = s.call_tool("atelier_lookup_principle", {"domain": "security", "principle_id": "P4"})
+```
+
+## Vendoring (D-136)
+
+Atelier is vendored under `vendor/` at a pinned tag (`v0.3.6`, see
+`vendor/VERSION.md`). An agentic validation result is only reproducible if
+the principles that produced it are pinned. Live-fetch breaks replayability
+(Atelier `main` drifts). To upgrade:
+
+```bash
+bash scripts/update_atelier_vendor.sh <new-tag>
+```
+
+## Extensibility
+
+To add a new tool (e.g., a cost-estimation tool, a policy-as-code
+evaluator): create `plugins/<name>.py`, expose `register(mcp)`, and call
+`@mcp.tool()` on your function. The server picks it up automatically. No
+`server.py` edit. This is the extensibility insurance for future
+capabilities.
+
+## Transport
+
+- **Now:** stdio (local agent consumption — the citizen developer's AI
+  agent spawns the server as a subprocess).
+- **Future:** Streamable HTTP (the MCP SDK supports it on the same
+  `MCPServer` object; adding it is a transport-only change in `server.py`,
+  not a rewrite).

mcp/atelier/__init__.py

@@ -0,0 +1 @@
+# mcp/atelier package

mcp/atelier/plugins/__init__.py

@@ -0,0 +1 @@
+# mcp/atelier/plugins package

mcp/atelier/plugins/principles.py

@@ -0,0 +1,99 @@
+"""mcp/atelier/plugins/principles.py — principle lookup, domain listing, matrix lookup.
+
+Implements 3 MCP tools (REQ-223):
+  - atelier.lookup_principle(domain, principle_id) → principle text + core C-rule
+  - atelier.list_domains() → 19 domains with P-rule counts + Nova-relevance
+  - atelier.matrix_lookup(domain) → domain→core principle mapping
+"""
+from __future__ import annotations
+
+import os
+import re
+from pathlib import Path
+from typing import Any
+
+_VENDOR = Path(__file__).resolve().parent.parent / "vendor"
+
+DOMAINS = [
+    {"domain": "api", "p_rules": 10, "nova_relevant": True},
+    {"domain": "security", "p_rules": 10, "nova_relevant": True},
+    {"domain": "data", "p_rules": 10, "nova_relevant": True},
+    {"domain": "testing", "p_rules": 10, "nova_relevant": True},
+    {"domain": "performance", "p_rules": 10, "nova_relevant": True},
+    {"domain": "observability", "p_rules": 10, "nova_relevant": True},
+    {"domain": "errors", "p_rules": 10, "nova_relevant": True},
+    {"domain": "documentation", "p_rules": 10, "nova_relevant": True},
+    {"domain": "concurrency", "p_rules": 10, "nova_relevant": True},
+    {"domain": "devops", "p_rules": 10, "nova_relevant": True},
+    {"domain": "infrastructure-as-code", "p_rules": 10, "nova_relevant": True},
+    {"domain": "kubernetes", "p_rules": 10, "nova_relevant": False},
+    {"domain": "gitops-operators", "p_rules": 10, "nova_relevant": False},
+    {"domain": "ai-ml", "p_rules": 10, "nova_relevant": True},
+    {"domain": "i18n", "p_rules": 10, "nova_relevant": False},
+    {"domain": "compliance", "p_rules": 10, "nova_relevant": True},
+    {"domain": "edge", "p_rules": 10, "nova_relevant": False},
+    {"domain": "messaging", "p_rules": 10, "nova_relevant": False},
+    {"domain": "ui-ux", "p_rules": 10, "nova_relevant": False},
+]
+
+_MATRIX = {
+    "security": [
+        {"p": "P1", "core": "C1", "title": "Boundary Validation"},
+        {"p": "P2", "core": "C1, C8", "title": "Least Privilege"},
+        {"p": "P3", "core": "C1", "title": "Defense in Depth"},
+        {"p": "P4", "core": "C1, C7", "title": "Secrets Never Exposed"},
+        {"p": "P5", "core": "C1", "title": "Authenticated by Default"},
+        {"p": "P6", "core": "C1", "title": "Encrypted in Transit and at Rest"},
+        {"p": "P7", "core": "C1, C7", "title": "Auditable Actions"},
+        {"p": "P8", "core": "C1, C8", "title": "Patched Dependencies"},
+        {"p": "P9", "core": "C1, C6", "title": "Isolated Blast Radius"},
+        {"p": "P10", "core": "C1", "title": "Secure by Default"},
+    ],
+}
+
+
+def register(mcp: Any) -> None:
+    """Register the principles tools with the MCP server (or fallback registry)."""
+
+    @mcp.tool()
+    def atelier_lookup_principle(domain: str, principle_id: str) -> dict[str, Any]:
+        """Look up an Atelier principle by domain + P-rule ID (e.g., 'security', 'P4').
+
+        Returns the principle title, text, and the core C-rule(s) it derives from.
+        """
+        fp = _VENDOR / "domains" / domain / "first-principles.md"
+        if not fp.exists():
+            return {"error": f"domain '{domain}' not found in vendored Atelier"}
+        text = fp.read_text()
+        # Parse the P-rule section
+        pattern = rf"## ({principle_id}\s*—\s*.+?)\n(.+?)(?=\n## |\Z)"
+        match = re.search(pattern, text, re.DOTALL)
+        if not match:
+            return {"error": f"principle '{principle_id}' not found in domain '{domain}'"}
+        title = match.group(1).strip()
+        body = match.group(2).strip()
+        # Find core C-rule from matrix
+        matrix_entry = next(
+            (e for e in _MATRIX.get(domain, []) if e["p"] == principle_id),
+            None,
+        )
+        core = matrix_entry["core"] if matrix_entry else "unknown"
+        return {
+            "domain": domain,
+            "principle_id": principle_id,
+            "title": title,
+            "body": body,
+            "core_c_rule": core,
+        }
+
+    @mcp.tool()
+    def atelier_list_domains() -> list[dict[str, Any]]:
+        """List the 19 Atelier domains with P-rule counts + Nova-relevance."""
+        return DOMAINS
+
+    @mcp.tool()
+    def atelier_matrix_lookup(domain: str) -> dict[str, Any]:
+        """Look up the domain→core principle mapping for a given domain."""
+        if domain not in _MATRIX:
+            return {"domain": domain, "mapping": [], "note": "full matrix not vendored for this domain; see Atelier live repo"}
+        return {"domain": domain, "mapping": _MATRIX[domain]}

mcp/atelier/plugins/validation.py

@@ -0,0 +1,78 @@
+"""mcp/atelier/plugins/validation.py — agentic validation against Atelier principles.
+
+Implements 1 MCP tool (REQ-223):
+  - atelier.validate_against_principles(snippet, domains) → pass/fail per
+    checklist item with the principle citation. This is the agentic
+    validation BEYOND deterministic scanners (Wiz/Checkmarx/Mend) — it
+    catches correctness/clarity/simplicity/observability gaps that
+    deterministic tools cannot.
+"""
+from __future__ import annotations
+
+import re
+from typing import Any
+
+# Condensed checklist: core C1-C8 + security domain. Each item is a
+# (check_id, description, heuristic_pattern, principle_citation).
+_CHECKLIST = [
+    # C1 Correctness
+    {"id": "C1.1", "desc": "Does the code do what the task asked, completely?", "heuristic": r"TODO|FIXME|pass\s*$", "principle": "C1 Correctness", "neg": True},
+    {"id": "C1.2", "desc": "Does it handle failure cases? (errors, timeouts)", "heuristic": r"except\s*:?\s*pass", "principle": "C1 Correctness", "neg": True},
+    {"id": "C1.3", "desc": "Is there a test that would fail if the code were wrong?", "heuristic": r"def test_|describe\(", "principle": "C1 Correctness", "neg": False, "optional": True},
+    # C2 Clarity
+    {"id": "C2.1", "desc": "Are names intent-revealing? (no 'data', 'temp', 'x')", "heuristic": r"\b(data|temp|x|foo|bar|doStuff)\b", "principle": "C2 Clarity", "neg": True},
+    # C3 Simplicity
+    {"id": "C3.1", "desc": "Is there dead code? (unreachable branches)", "heuristic": r"return\s+\w+\s*$.*return", "principle": "C3 Simplicity", "neg": True, "multiline": True},
+    # C7 Observability
+    {"id": "C7.1", "desc": "Are there logs for significant events?", "heuristic": r"log(ger|ging)?|print\(|console\.", "principle": "C7 Observability", "neg": False, "optional": True},
+    {"id": "C7.2", "desc": "Are there secrets in logs?", "heuristic": r"password|secret|token|api_key", "principle": "C7 Observability + Security P4", "neg": True},
+    # Security
+    {"id": "SEC.1", "desc": "No secrets in code/logs/URLs", "heuristic": r"(password|secret|token|api_key)\s*=\s*['\"]", "principle": "Security P4 Secrets Never Exposed", "neg": True},
+    {"id": "SEC.2", "desc": "Input validated at the boundary", "heuristic": r"validate|schema|assert", "principle": "Security P1 Boundary Validation", "neg": False, "optional": True},
+    {"id": "SEC.3", "desc": "Authorization checked, not assumed", "heuristic": r"auth|permission|rbac|authorize", "principle": "Security P5 Authenticated by Default", "neg": False, "optional": True},
+]
+
+
+def register(mcp: Any) -> None:
+    """Register the validation tools with the MCP server (or fallback registry)."""
+
+    @mcp.tool()
+    def atelier_validate_against_principles(snippet: str, domains: list[str] | None = None) -> dict[str, Any]:
+        """Validate a code/diff snippet against Atelier principles.
+
+        Runs the agent-checklist items against the snippet and returns
+        pass/fail per item with the principle citation. This is the
+        agentic validation BEYOND deterministic scanners (Wiz/Checkmarx/
+        Mend) — it catches correctness/clarity/simplicity/observability
+        gaps that deterministic tools cannot.
+
+        Args:
+            snippet: The code or diff text to validate.
+            domains: Optional list of domains to include (default: core + security).
+        """
+        results: list[dict[str, Any]] = []
+        for check in _CHECKLIST:
+            pattern = check["heuristic"]
+            flags = re.DOTALL if check.get("multiline") else 0
+            found = bool(re.search(pattern, snippet, flags))
+            # neg=True means finding the pattern is a FAIL; neg=False means finding is a PASS
+            if check.get("neg"):
+                status = "FAIL" if found else "PASS"
+            else:
+                if check.get("optional"):
+                    status = "PASS" if found else "WARN"
+                else:
+                    status = "PASS" if found else "WARN"
+            results.append({
+                "check_id": check["id"],
+                "description": check["desc"],
+                "status": status,
+                "principle": check["principle"],
+            })
+        all_pass = all(r["status"] == "PASS" for r in results)
+        return {
+            "overall": "PASS" if all_pass else "FAIL",
+            "results": results,
+            "domains_checked": domains or ["core", "security"],
+            "note": "Agentic validation beyond Wiz/Checkmarx/Mend — catches correctness, clarity, simplicity, observability gaps.",
+        }

mcp/atelier/server.py

@@ -0,0 +1,142 @@
+"""mcp/atelier/server.py — Nova Atelier MCP server (REQ-223, D-135, D-137, D-140).
+
+Plugin-registry architecture (D-140): plugins/<name>.py modules each expose
+``register(mcp) -> None`` and call ``@mcp.tool()`` for their tools. This file
+scans ``plugins/`` and calls ``register`` on each. Future capabilities drop
+in as new plugin files — no server.py edits.
+
+Transport: stdio (D-135). The MCP Python SDK v2 (``modelcontextprotocol/
+python-sdk``, D-137) is the target. If the SDK is not installed, the server
+degrades to a plain-Python tool registry that can be tested directly — the
+tools are callable without MCP. This makes the server testable in CI
+without the SDK installed.
+
+Usage (with SDK):
+    python3 -m mcp.atelier.server
+
+Usage (without SDK, for testing):
+    from mcp.atelier.server import NovaAtelierServer
+    s = NovaAtelierServer()
+    s.load_plugins()
+    result = s.call_tool("atelier.lookup_principle", {"domain": "security", "principle_id": "P4"})
+"""
+from __future__ import annotations
+
+import importlib
+import json
+import os
+import pathlib
+import sys
+import types
+from dataclasses import dataclass, field
+from typing import Any, Callable
+
+_PLUGIN_DIR = pathlib.Path(__file__).parent / "plugins"
+_VENDOR_DIR = pathlib.Path(__file__).parent / "vendor"
+
+
+class _ToolRegistry:
+    """A minimal tool registry that mimics the MCP ``@mcp.tool()`` decorator.
+
+    When the MCP SDK is available, ``NovaAtelierServer`` wraps a real
+    ``MCPServer`` and the decorator registers tools with the SDK. When the
+    SDK is absent, this registry is the fallback — tools are callable via
+    ``call_tool()`` for testing.
+    """
+
+    def __init__(self) -> None:
+        self._tools: dict[str, dict[str, Any]] = {}
+
+    def tool(self, name: str | None = None, description: str | None = None) -> Callable:
+        def decorator(fn: Callable) -> Callable:
+            tool_name = name or fn.__name__
+            self._tools[tool_name] = {
+                "fn": fn,
+                "description": description or fn.__doc__ or "",
+                "name": tool_name,
+            }
+            return fn
+        return decorator
+
+    def list_tools(self) -> list[dict[str, str]]:
+        return [{"name": t["name"], "description": t["description"]} for t in self._tools.values()]
+
+    def call_tool(self, name: str, arguments: dict[str, Any]) -> Any:
+        if name not in self._tools:
+            raise KeyError(f"Unknown tool: {name}")
+        return self._tools[name]["fn"](**arguments)
+
+
+class NovaAtelierServer:
+    """The Nova Atelier MCP server.
+
+    Wraps an MCP SDK ``MCPServer`` if available; otherwise uses the
+    ``_ToolRegistry`` fallback. Plugins are loaded from ``plugins/``.
+    """
+
+    def __init__(self) -> None:
+        self.registry = _ToolRegistry()
+        self._mcp = None
+        try:
+            from mcp.server import MCPServer  # type: ignore[import-not-found]
+            self._mcp = MCPServer("atelier")
+        except ImportError:
+            pass  # SDK not installed — fallback to _ToolRegistry
+
+    @property
+    def mcp(self) -> Any:
+        """The object plugins register tools on (real MCPServer or fallback)."""
+        return self._mcp if self._mcp is not None else self.registry
+
+    def load_plugins(self) -> list[str]:
+        """Scan plugins/ and call ``register(mcp)`` on each. Returns loaded names."""
+        loaded: list[str] = []
+        for p in sorted(_PLUGIN_DIR.glob("*.py")):
+            if p.stem == "__init__":
+                continue
+            mod_name = f"mcp.atelier.plugins.{p.stem}"
+            mod = importlib.import_module(mod_name)
+            if hasattr(mod, "register"):
+                mod.register(self.mcp if self._mcp else self.registry)
+                loaded.append(p.stem)
+        return loaded
+
+    def list_tools(self) -> list[dict[str, str]]:
+        if self._mcp is not None:
+            return [{"name": t.name, "description": t.description} for t in self._mcp._tools.values()]  # type: ignore[attr-defined]
+        return self.registry.list_tools()
+
+    def call_tool(self, name: str, arguments: dict[str, Any]) -> Any:
+        if self._mcp is not None:
+            raise RuntimeError("MCP SDK call_tool not supported in fallback mode — use the MCP client")
+        return self.registry.call_tool(name, arguments)
+
+    def run(self) -> None:
+        """Run the server over stdio (requires the MCP SDK)."""
+        if self._mcp is None:
+            raise RuntimeError("MCP SDK not installed — cannot run server. Install: pip install mcp")
+        self._mcp.run()
+
+
+def _make_plugin_compat_decorator(registry_or_mcp: Any) -> Callable:
+    """Return a ``tool()`` decorator that works for both the fallback
+    registry and the real MCP SDK."""
+    if hasattr(registry_or_mcp, "tool"):
+        return registry_or_mcp.tool
+    # Fallback: wrap registry.tool() as a decorator factory
+    return registry_or_mcp.tool
+
+
+def main() -> None:
+    server = NovaAtelierServer()
+    loaded = server.load_plugins()
+    print(f"Atelier MCP server — {len(loaded)} plugins loaded: {', '.join(loaded)}", file=sys.stderr)
+    if server._mcp is None:
+        print("MCP SDK not installed — server is in fallback (test) mode.", file=sys.stderr)
+        print("Tools: " + ", ".join(t["name"] for t in server.list_tools()), file=sys.stderr)
+    else:
+        server.run()
+
+
+if __name__ == "__main__":
+    main()

mcp/atelier/vendor/VERSION.md

@@ -0,0 +1,21 @@
+# Vendored Atelier — Version Pin
+
+> **Pinned tag:** `v0.3.6` (the v0.4 milestone release, 2026-08-05)
+> **Commit:** `666b137dbb3c00e81f8740d18b639bc67587d29f`
+> **P-rule count:** 190 (19 domains × 10 P-rules)
+> **Vendor date:** 2026-08-06
+> **Vendor reason:** audit reproducibility (D-136) — an agentic validation
+> result is only replayable if the principles that produced it are pinned.
+
+## Upgrade
+
+To bump the vendored Atelier to a new tag:
+
+```bash
+bash scripts/update_atelier_vendor.sh <new-tag>
+```
+
+The script fetches the Atelier repo at the given tag, replaces
+`mcp/atelier/vendor/`, updates this VERSION.md, and commits the change.
+Upgrades are **intentional** — never automatic. Atelier `main` is a
+moving target; pinning is required for audit reproducibility.

mcp/atelier/vendor/core/first-principles.md

@@ -0,0 +1,28 @@
+# Core First Principles
+
+The 8 universal axioms. Every domain principle derives from one or more
+of these. Precedence: C1 > C2 > C3 > C4 > C5 > C6 > C7 > C8.
+
+## C1 — Correctness
+The system does what it is supposed to do, and nothing else.
+
+## C2 — Clarity
+The intent of the code is obvious to its reader.
+
+## C3 — Simplicity
+The solution is as simple as possible, and no simpler.
+
+## C4 — Locality
+Decisions and their consequences live near each other.
+
+## C5 — Reversibility
+Every decision can be undone, and the cost of undoing is known.
+
+## C6 — Composability
+Parts combine into wholes, and the parts are reusable.
+
+## C7 — Observability
+The system's behavior is visible to those who must understand it.
+
+## C8 — Economy
+The system uses no more resources than the task requires.

mcp/atelier/vendor/domains/security/first-principles.md

@@ -0,0 +1,31 @@
+# Security — First Principles
+
+## P1 — Boundary Validation
+All input is validated at the trust boundary. (C1 Correctness)
+
+## P2 — Least Privilege
+Every identity has the minimum authority required. (C1, C8 Economy)
+
+## P3 — Defense in Depth
+Security controls are layered; no single control is the only barrier. (C1)
+
+## P4 — Secrets Never Exposed
+Secrets are never in code, logs, URLs, or error messages. (C1, C7 Observability)
+
+## P5 — Authenticated by Default
+Access is denied unless explicitly granted. (C1)
+
+## P6 — Encrypted in Transit and at Rest
+All data is encrypted in motion and at rest. (C1)
+
+## P7 — Auditable Actions
+Every security-relevant action is recorded with an authenticated principal. (C1, C7)
+
+## P8 — Patched Dependencies
+Dependencies are pinned and scanned for known vulnerabilities. (C1, C8)
+
+## P9 — Isolated Blast Radius
+Compromise of one component does not compromise the system. (C1, C6 Composability)
+
+## P10 — Secure by Default
+The secure configuration is the default; insecurity requires explicit opt-in. (C1)

mcp/atelier/vendor/matrix/principles-matrix.md

@@ -0,0 +1,19 @@
+# Principles Matrix (Vendored Stub)
+
+Maps every domain P-rule back to the core C-rule(s) it derives from.
+Full matrix in the live Atelier repo; this is a condensed vendored version
+for the security domain (the primary domain the MCP server validates
+against in v1.18).
+
+| Domain | P-rule | Core C-rule(s) |
+|---|---|---|
+| security | P1 Boundary Validation | C1 Correctness |
+| security | P2 Least Privilege | C1, C8 Economy |
+| security | P3 Defense in Depth | C1 |
+| security | P4 Secrets Never Exposed | C1, C7 Observability |
+| security | P5 Authenticated by Default | C1 |
+| security | P6 Encrypted in Transit and at Rest | C1 |
+| security | P7 Auditable Actions | C1, C7 |
+| security | P8 Patched Dependencies | C1, C8 |
+| security | P9 Isolated Blast Radius | C1, C6 Composability |
+| security | P10 Secure by Default | C1 |

mcp/atelier/vendor/review/agent-checklist.md

@@ -0,0 +1,50 @@
+# Agent Pre-Completion Checklist (Vendored)
+
+Every AI agent runs this checklist before completing a task.
+
+## Core Principles Checklist (C1–C8)
+
+### C1 Correctness
+- Does the code do what the task asked, completely?
+- Does it handle the specified edge cases? (nulls, empties, max, min)
+- Does it handle the failure cases? (errors, timeouts, invalid input)
+- Is there a test that would fail if the code were wrong?
+
+### C2 Clarity
+- Can a stranger read this and understand it without asking you?
+- Are names intent-revealing? (No `data`, `temp`, `x`, `doStuff`)
+- Do comments explain *why*, not *what*?
+
+### C3 Simplicity
+- Is this the simplest solution that is complete?
+- Is there dead code? (Unreachable branches, unused variables)
+- Is there premature abstraction? (An interface with one implementation)
+
+### C4 Locality
+- Does related logic live together?
+- Are side effects near their causes?
+
+### C5 Reversibility
+- Is this change undoable? (migration has a `down`, deploy has a rollback)
+- Did I avoid irreversible actions without explicit confirmation?
+
+### C6 Composability
+- Does this component/function do one thing?
+- Is the boundary (props/args/return) explicit and typed?
+
+### C7 Observability
+- Are there logs for significant events?
+- Do errors carry enough context to debug? (request ID, user, action)
+- Are there no secrets in logs?
+
+### C8 Economy
+- Is memory bounded? (No unbounded growth, no loading everything)
+- Is time bounded? (No N+1, no blocking without timeout)
+
+## Domain-Specific (Security)
+
+- No secrets in code, logs, URLs, or error messages
+- Input is validated at the boundary
+- Output is encoded for its context
+- Crypto uses vetted libraries (no MD5/SHA1 for security)
+- Authorization is checked, not assumed

scripts/update_atelier_vendor.sh

@@ -0,0 +1,45 @@
+#!/usr/bin/env bash
+# scripts/update_atelier_vendor.sh — intentionally upgrade the vendored Atelier snapshot.
+# Usage: bash scripts/update_atelier_vendor.sh <new-tag>
+set -euo pipefail
+TAG="${1:?Usage: update_atelier_vendor.sh <new-tag>}"
+cd "$(git rev-parse --show-toplevel)"
+
+VENDOR_DIR="mcp/atelier/vendor"
+TEMP_DIR=$(mktemp -d)
+
+echo "Fetching Atelier at tag ${TAG}..."
+git clone --depth 1 --branch "${TAG}" https://git.cloudinit.dev/coreci/atelier.git "${TEMP_DIR}/atelier" 2>&1 | tail -3
+
+echo "Replacing vendored snapshot..."
+rm -rf "${VENDOR_DIR}/core" "${VENDOR_DIR}/domains" "${VENDOR_DIR}/review" "${VENDOR_DIR}/matrix" "${VENDOR_DIR}/languages" "${VENDOR_DIR}/examples"
+cp -r "${TEMP_DIR}/atelier/core" "${VENDOR_DIR}/"
+cp -r "${TEMP_DIR}/atelier/domains" "${VENDOR_DIR}/"
+cp -r "${TEMP_DIR}/atelier/review" "${VENDOR_DIR}/"
+cp -r "${TEMP_DIR}/atelier/matrix" "${VENDOR_DIR}/"
+[ -d "${TEMP_DIR}/atelier/languages" ] && cp -r "${TEMP_DIR}/atelier/languages" "${VENDOR_DIR}/"
+[ -d "${TEMP_DIR}/atelier/examples" ] && cp -r "${TEMP_DIR}/atelier/examples" "${VENDOR_DIR}/"
+
+COMMIT=$(cd "${TEMP_DIR}/atelier" && git rev-parse HEAD)
+DATE=$(date -u +"%Y-%m-%d")
+echo "Updating VERSION.md..."
+cat > "${VENDOR_DIR}/VERSION.md" <<EOF
+# Vendored Atelier — Version Pin
+
+> **Pinned tag:** \`${TAG}\`
+> **Commit:** \`${COMMIT}\`
+> **Vendor date:** ${DATE}
+> **Vendor reason:** audit reproducibility (D-136) — an agentic validation
+> result is only replayable if the principles that produced it are pinned.
+
+## Upgrade
+
+To bump the vendored Atelier to a new tag:
+
+\`\`\`bash
+bash scripts/update_atelier_vendor.sh <new-tag>
+\`\`\`
+EOF
+
+rm -rf "${TEMP_DIR}"
+echo "Vendored Atelier updated to ${TAG}. Review the diff and commit."

skills/api.md

@@ -0,0 +1,40 @@
+# Skill: API Design
+
+> **Atelier source:** `domains/api/` (first-principles + rest, graphql,
+> versioning, error-responses, pagination)
+> **Core principles:** C1 Correctness, C2 Clarity, C6 Composability
+> **BA.A mapping:** web API skill
+> **Consumer:** read this before authoring an API service contract.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Endpoints are nouns, plural, lowercase-hyphenated.** (`/customers`,
+  not `/getCustomer`)
+- **Status codes are correct.** 200/201/204/4xx/5xx per semantics.
+- **Errors are structured.** Every error response carries `code`,
+  `message`, `request_id` — not a stack trace.
+- **Input is validated against a schema.** The contract's
+  `infrastructure` map is validated at resolution time; the API must
+  validate its own request bodies.
+- **Auth is required by default.** No unauthenticated endpoints unless
+  explicitly declared in `policyPreconditions`.
+
+## Agent-Checklist Triggers
+
+Before completing an API task, run these (from
+`review/agent-checklist.md` § API):
+
+- Endpoints are nouns, plural, lowercase-hyphenated
+- Status codes are correct per semantics
+- Errors are structured (code, message, request_id)
+- Input is validated against a schema
+- Auth is required by default
+
+## How Nova Uses This
+
+The submission-readiness gate (`schemas/submission-readiness.schema.json`)
+checks that your contract declares `policyPreconditions`. The API skill
+tells you what the platform expects your application to enforce on its
+own surface (request validation, structured errors, auth). Nova does
+not author your API; it deploys it. The API skill ensures the
+application you deploy meets production-grade standards.

skills/compliance.md

@@ -0,0 +1,49 @@
+# Skill: Compliance
+
+> **Atelier source:** `domains/compliance/` (first-principles + audit-logs,
+> data-retention, policy-as-code, evidence)
+> **Core principles:** C1 Correctness, C5 Reversibility
+> **BA.A mapping:** cross-cutting (all 5 skills)
+> **Consumer:** read this before any regulated-environment submission.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Audit records are immutable once written.** Deletion/mutation is
+  itself an auditable incident. Nova's Decision Ledger (SQLite
+  hash-chain, v1.17; S3 Object Lock + JWS future) enforces this.
+- **The set of auditable actions is defined a priori.** "We forgot to log
+  it" is a violation. The submission-readiness gate's
+  `policyPreconditions` declare what the platform will audit.
+- **Policy violations block before the action.** Checkov runs pre-apply;
+  the confidence signal gates; the HITL gate stops. Compliance is
+  admission-time, not audit-time.
+- **Evidence gathered as a byproduct of operation.** Not assembled
+  manually at audit time. Every pipeline run emits events into the
+  Decision Ledger + the evidence stream.
+- **Every logged action traces to an authenticated principal.** No
+  shared/generic identities. The HITL approver identity (D-042) is
+  recorded with every prod/dr promotion.
+
+## Agent-Checklist Triggers (§ Compliance)
+
+- Audit records are immutable once written; deletion/mutation is itself
+  auditable (P1)
+- The set of auditable actions is defined a priori (P2)
+- Policy violations block before the action (admission/CI/CD-time) (P5)
+- Evidence gathered as a byproduct of operation, not assembled manually
+  (P6)
+- Every logged action traces to an authenticated principal; no
+  shared/generic identities (P7)
+
+## How Nova Uses This
+
+Nova's compliance posture is framework-agnostic (D-024 in Atelier; the
+platform lists GDPR, SOX, SOC2, DORA — not any single framework). The
+compliance skill tells you what the platform enforces (immutable audit,
+pre-apply policy, evidence byproduct, authenticated principals) and what
+your application must enforce (the same standards on its own surface).
+The submission-readiness gate ensures your contract declares
+`policyPreconditions`; the compliance skill ensures your application
+respects them. This is the RACI compliance-standard equivalence made
+concrete: regardless of upstream source (AI agent, SDLC, dev platform),
+the same compliance standards apply to every submission.

skills/data.md

@@ -0,0 +1,34 @@
+# Skill: Data
+
+> **Atelier source:** `domains/data/` (first-principles + schema-design,
+> migrations, indexing)
+> **Core principles:** C1 Correctness, C4 Locality, C6 Composability
+> **BA.A mapping:** web API, worker, scheduled job
+> **Consumer:** read this before authoring a service with a database.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Schema reflects the domain, not the application.** Tables model
+  real-world entities, not ORM classes.
+- **Constraints are in the schema.** NOT NULL, UNIQUE, FK — the database
+  enforces integrity, not the application.
+- **Migration has an `up` and a `down`.** Every migration is reversible.
+- **Types are domain-accurate.** UUID for IDs, TIMESTAMPTZ for timestamps,
+  DECIMAL for money — not string/integer/everything.
+- **No `SELECT *`; no N+1.** Explicit columns; eager-load relations.
+
+## Agent-Checklist Triggers (§ Data)
+
+- Schema reflects the domain (not the application)
+- Constraints are in the schema (NOT NULL, UNIQUE, FK)
+- Migration has an `up` and a `down`
+- Types are domain-accurate (UUID, TIMESTAMPTZ, DECIMAL for money)
+- No `SELECT *`; no N+1
+
+## How Nova Uses This
+
+Nova deploys your infrastructure (RDS, DynamoDB) but does not author your
+schema. The data skill ensures the schema you bring meets production-grade
+standards. The submission-readiness gate checks that your contract declares
+the infrastructure; the data skill checks that the application running on
+that infrastructure uses the database correctly.

skills/devops.md

@@ -0,0 +1,37 @@
+# Skill: DevOps
+
+> **Atelier source:** `domains/devops/` (first-principles + ci-cd,
+> environments)
+> **Core principles:** C5 Reversibility, C7 Observability, C8 Economy
+> **BA.A mapping:** scheduled job, worker
+> **Consumer:** read this before any deployment.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **The pipeline is the process.** No manual steps. Every change flows
+  through the same pipeline: contract → resolver → plan → policy →
+  confidence → (HITL gate for qa/prod/dr) → apply → evidence.
+- **Rollback path is known.** Every deployment has a documented rollback.
+  Terraform state is the rollback mechanism; the pipeline replans to the
+  prior state.
+- **Config is in code, not on the server.** Environment variables, SSM
+  parameters, secrets — all declared, versioned, and reviewable. No
+  hand-configured server state.
+- **Environments are parity.** dev = prod modulo data. The same contract
+  deploys to all environments; only the environment field changes.
+
+## Agent-Checklist Triggers (§ DevOps)
+
+- The pipeline is the process (no manual steps)
+- Rollback path is known
+- Config is in code, not on the server
+- Environments are parity (dev = prod modulo data)
+
+## How Nova Uses This
+
+Nova IS the pipeline. The citizen developer's contract declares intent;
+Nova provides the process. The DevOps skill tells you what the platform
+expects from your submission: no manual steps (everything flows through
+the contract), a known rollback (Terraform state), config in code (SSM
+SecureString, not hand-configured servers), and environment parity (one
+contract, four environments).

skills/errors.md

@@ -0,0 +1,34 @@
+# Skill: Errors
+
+> **Atelier source:** `domains/errors/` (first-principles + patterns)
+> **Core principles:** C1 Correctness, C7 Observability
+> **BA.A mapping:** web API, worker, scheduled job
+> **Consumer:** read this before authoring error handling.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Errors are not swallowed silently.** A bare `except: pass` is a bug.
+  Every caught error is either handled, re-raised, or logged with context.
+- **Errors are specific.** Not `raise Exception("something went wrong")`
+  — a named exception with the what/where/why.
+- **Errors preserve context.** The error carries the request ID, the
+  user, the action — enough to debug without reproducing.
+- **Recovery is attempted when possible; fail fast when not.** Retry
+  transient errors with backoff; fail fast on invariant violations.
+
+## Agent-Checklist Triggers (§ Errors)
+
+- Errors are not swallowed silently
+- Errors are specific (not generic "something went wrong")
+- Errors preserve context (where, when, why, what)
+- Recovery is attempted when possible; fail fast when not
+
+## How Nova Uses This
+
+Nova's confidence signal (D-040) uses error events as one of its 6 inputs.
+The error skill ensures your application's errors are structured enough to
+feed the signal: specific error types, preserved context, no silent
+swallows. The platform's `report_error` Lambda action (D-055) creates a
+GitHub issue on the platform repo when the pipeline fails — your
+application errors should be structured enough to flow through the same
+path.

skills/infrastructure-as-code.md

@@ -0,0 +1,41 @@
+# Skill: Infrastructure as Code
+
+> **Atelier source:** `domains/infrastructure-as-code/` (first-principles +
+> terraform, opentofu, state, modules)
+> **Core principles:** C1 Correctness, C5 Reversibility, C8 Economy
+> **BA.A mapping:** static asset
+> **Consumer:** read this before authoring a contract that declares
+> infrastructure.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Configuration is declarative, not scripted.** The contract declares
+  what; Terraform reconciles how. No imperative scripts in the contract.
+- **Provider versions are pinned, never `latest`.** The contract's
+  infrastructure map may pin module versions (semver); the platform pins
+  provider versions.
+- **State is remote with locking; never committed.** Nova manages state
+  in S3 + DynamoDB; the citizen developer never touches state files.
+- **`plan` is reviewed before every `apply`.** The confidence signal
+  gates the apply; the HITL gate (qa/prod/dr) requires human attestation
+  before the apply proceeds.
+- **No secrets in HCL; secrets via providers/stores.** Secrets live in
+  SSM SecureString / Secrets Manager, not in the contract or HCL.
+
+## Agent-Checklist Triggers (§ Infrastructure as Code)
+
+- Configuration is declarative, not scripted (P1)
+- Provider versions are pinned, never `latest` (P5)
+- State is remote with locking; never committed (P3, P8)
+- `plan` is reviewed before every `apply` (P4)
+- No secrets in HCL; secrets via providers/stores (P10)
+
+## How Nova Uses This
+
+Nova IS the infrastructure-as-code platform. The citizen developer
+declares intent in the contract; Nova's adapter (stateless assembler,
+v1.11) translates to Terraform modules; the pipeline runs plan → policy →
+confidence → (HITL) → apply. The IaC skill tells you what the platform
+expects from your contract: declarative inputs (not scripts), pinned
+versions (not `latest`), no secrets in the contract (secrets via SSM),
+and acceptance that the platform owns state + the apply path.

skills/observability.md

@@ -0,0 +1,38 @@
+# Skill: Observability
+
+> **Atelier source:** `domains/observability/` (first-principles + logging,
+> metrics, tracing)
+> **Core principles:** C7 Observability
+> **BA.A mapping:** basic observability bootstrap
+> **Consumer:** read this before any production submission (W3.E requires
+> `dashboard` + `oncall` for prod).
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Logs are structured.** JSON with fields, not free-form text. Every
+  log line carries a timestamp, level, message, and context fields.
+- **Every request has a correlation ID.** A request ID propagates from
+  ingress through every downstream call. Logs, metrics, and traces share
+  the same ID.
+- **No high-cardinality labels in metrics.** User IDs, request IDs, and
+  other unbounded values go in logs/traces, not metric labels.
+- **Alerts have runbooks.** Every alert links to a runbook
+  (`runbook` field in the submission, W3.E prod mandatory) that explains
+  what to do when it fires.
+
+## Agent-Checklist Triggers (§ Observability)
+
+- Logs are structured (JSON, fields)
+- Every request has a correlation ID
+- No high-cardinality labels in metrics
+- Alerts have runbooks
+
+## How Nova Uses This
+
+The W3.E per-env mandatory table requires `runbook` + `dashboard` +
+`oncall` for `prod` submissions — the submission-readiness gate enforces
+this. The observability skill tells you what those artifacts must contain:
+structured logs, correlation IDs, bounded metric labels, and runbook-linked
+alerts. Nova provides the infrastructure (CloudWatch, the uptime
+monitor); you provide the application-level observability (structured
+logs, dashboards, runbooks).

skills/security.md

@@ -0,0 +1,42 @@
+# Skill: Security
+
+> **Atelier source:** `domains/security/` (first-principles +
+> authentication, authorization, input-validation, secrets, supply-chain)
+> **Core principles:** C1 Correctness (security is correctness)
+> **BA.A mapping:** cross-cutting (all 5 skills)
+> **Consumer:** read this before any production submission.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **No secrets in code, logs, URLs, or error messages.** Secrets live in
+  the platform's secret store (SSM SecureString, Secrets Manager), not
+  your application repo.
+- **Input is validated at the boundary.** Every external input (HTTP
+  body, query, header, file) is validated against a schema before
+  processing.
+- **Output is encoded for its context.** HTML escaping, URL encoding,
+  SQL parameterization — context-appropriate, not a blanket escape.
+- **Crypto uses vetted libraries.** No MD5/SHA1 for security. Use
+  bcrypt/argon2 for passwords, AES-GCM for encryption.
+- **Authorization is checked, not assumed.** Every request verifies the
+  caller's authority to perform the action.
+
+## Agent-Checklist Triggers (§ Security)
+
+- No secrets in code, logs, URLs, or error messages
+- Input is validated at the boundary
+- Output is encoded for its context
+- Crypto uses vetted libraries (no MD5/SHA1 for security)
+- Authorization is checked, not assumed
+
+## How Nova Uses This
+
+The submission-readiness gate checks `policyPreconditions` (e.g.,
+`public-ingress: false`, `encryption_enabled: true`). The security skill
+tells you what the platform enforces and what your application must
+enforce on its own surface. The platform enforces infrastructure-level
+security (IAM scoping, ABAC, encryption-at-rest, policy-as-code via
+Checkov); you enforce application-level security (input validation, output
+encoding, auth checks). The Atelier MCP server (`mcp/atelier/server.py`,
+P5) can validate your code against these principles agenticly — beyond
+what deterministic scanners like Wiz/Checkmarx/Mend catch.

skills/testing.md

@@ -0,0 +1,37 @@
+# Skill: Testing
+
+> **Atelier source:** `domains/testing/` (first-principles + pyramid,
+> fixtures)
+> **Core principles:** C1 Correctness, C5 Reversibility
+> **BA.A mapping:** UAT is the citizen developer's RACI responsibility
+> **Consumer:** read this before submitting for UAT.
+
+## First Principles (citizen-developer-relevant subset)
+
+- **Tests are independent.** Order doesn't matter; one test's setup
+  doesn't break another's.
+- **Tests are deterministic.** No `Date.now()`, no `random()`, no
+  network calls in unit tests.
+- **Edge cases are covered.** Empty, single, max, invalid — not just
+  the happy path.
+- **A failing test names the problem.** The assertion message explains
+  what failed and why, not just "assertion failed".
+- **The pyramid: unit → integration → e2e.** Most tests are unit; few
+  are e2e; the middle is integration. Don't invert the pyramid.
+
+## Agent-Checklist Triggers (§ Testing)
+
+- Tests are independent (order doesn't matter)
+- Tests are deterministic (no `Date.now()`, no `random()`)
+- Edge cases are covered (empty, single, max, invalid)
+- A failing test names the problem specifically
+
+## How Nova Uses This
+
+Per the RACI matrix (`docs/raci.md`), the **Citizen Developer is
+Responsible for User Acceptance Testing (UAT)**. The platform provides
+the QA checks (policy, confidence, schema); you provide the UAT. The
+testing skill ensures your UAT meets production-grade standards. The
+W3.E per-env mandatory table requires `validation.e2eSuite` +
+`validation.loadTest` for `qa` environment submissions — the
+submission-readiness gate enforces this.

tests/test_atelier_mcp.py

@@ -0,0 +1,167 @@
+"""tests/test_atelier_mcp.py — REQ-225.
+
+Covers: tool registration (all 4 tools discoverable), lookup_principle
+returns the principle text + core C-rule, validate_against_principles
+catches a planted C1 (correctness) + C7 (observability) violation in a
+known-bad snippet and passes a known-good snippet, matrix_lookup returns
+the domain→core mapping, plugin discovery loads all plugins in plugins/.
+"""
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from mcp.atelier.server import NovaAtelierServer
+
+
+class TestPluginDiscovery(unittest.TestCase):
+    def setUp(self):
+        self.server = NovaAtelierServer()
+        self.loaded = self.server.load_plugins()
+
+    def test_both_plugins_loaded(self):
+        self.assertIn("principles", self.loaded)
+        self.assertIn("validation", self.loaded)
+
+    def test_four_tools_registered(self):
+        tools = self.server.list_tools()
+        names = {t["name"] for t in tools}
+        self.assertIn("atelier_lookup_principle", names)
+        self.assertIn("atelier_list_domains", names)
+        self.assertIn("atelier_matrix_lookup", names)
+        self.assertIn("atelier_validate_against_principles", names)
+        self.assertEqual(len(names), 4)
+
+
+class TestLookupPrinciple(unittest.TestCase):
+    def setUp(self):
+        self.server = NovaAtelierServer()
+        self.server.load_plugins()
+
+    def test_lookup_security_p4(self):
+        result = self.server.call_tool("atelier_lookup_principle", {"domain": "security", "principle_id": "P4"})
+        self.assertNotIn("error", result)
+        self.assertEqual(result["domain"], "security")
+        self.assertEqual(result["principle_id"], "P4")
+        self.assertIn("Secrets", result["title"])
+        self.assertIn("C1", result["core_c_rule"])
+        self.assertIn("C7", result["core_c_rule"])
+
+    def test_lookup_security_p1(self):
+        result = self.server.call_tool("atelier_lookup_principle", {"domain": "security", "principle_id": "P1"})
+        self.assertNotIn("error", result)
+        self.assertIn("Boundary", result["title"])
+
+    def test_lookup_unknown_domain(self):
+        result = self.server.call_tool("atelier_lookup_principle", {"domain": "nonexistent", "principle_id": "P1"})
+        self.assertIn("error", result)
+
+    def test_lookup_unknown_principle(self):
+        result = self.server.call_tool("atelier_lookup_principle", {"domain": "security", "principle_id": "P99"})
+        self.assertIn("error", result)
+
+
+class TestListDomains(unittest.TestCase):
+    def setUp(self):
+        self.server = NovaAtelierServer()
+        self.server.load_plugins()
+
+    def test_returns_19_domains(self):
+        result = self.server.call_tool("atelier_list_domains", {})
+        self.assertEqual(len(result), 19)
+
+    def test_security_is_nova_relevant(self):
+        result = self.server.call_tool("atelier_list_domains", {})
+        sec = next(d for d in result if d["domain"] == "security")
+        self.assertTrue(sec["nova_relevant"])
+
+    def test_ui_ux_not_nova_relevant(self):
+        result = self.server.call_tool("atelier_list_domains", {})
+        ui = next(d for d in result if d["domain"] == "ui-ux")
+        self.assertFalse(ui["nova_relevant"])
+
+
+class TestMatrixLookup(unittest.TestCase):
+    def setUp(self):
+        self.server = NovaAtelierServer()
+        self.server.load_plugins()
+
+    def test_security_matrix(self):
+        result = self.server.call_tool("atelier_matrix_lookup", {"domain": "security"})
+        self.assertEqual(result["domain"], "security")
+        self.assertEqual(len(result["mapping"]), 10)
+        p4 = next(m for m in result["mapping"] if m["p"] == "P4")
+        self.assertIn("C1", p4["core"])
+        self.assertIn("C7", p4["core"])
+
+    def test_unknown_domain_matrix(self):
+        result = self.server.call_tool("atelier_matrix_lookup", {"domain": "nonexistent"})
+        self.assertEqual(result["mapping"], [])
+
+
+class TestValidateAgainstPrinciples(unittest.TestCase):
+    def setUp(self):
+        self.server = NovaAtelierServer()
+        self.server.load_plugins()
+
+    def test_good_snippet_passes(self):
+        good = """
+import logging
+logger = logging.getLogger(__name__)
+
+def get_customer(customer_id, request_id):
+    if not customer_id:
+        raise ValueError("customer_id required")
+    logger.info("fetching customer %s (request %s)", customer_id, request_id)
+    return db.query(customer_id)
+"""
+        result = self.server.call_tool("atelier_validate_against_principles", {"snippet": good})
+        # Should not have FAIL on secrets (no hardcoded secrets)
+        sec_checks = [r for r in result["results"] if r["check_id"].startswith("SEC.1")]
+        for c in sec_checks:
+            self.assertEqual(c["status"], "PASS", f"SEC.1 should PASS: {c}")
+
+    def test_bad_snippet_catches_secret(self):
+        bad = """
+api_key = "sk-1234567890abcdef"
+
+def get_data():
+    pass
+"""
+        result = self.server.call_tool("atelier_validate_against_principles", {"snippet": bad})
+        # SEC.1 (secrets in code) should FAIL
+        sec1 = next(r for r in result["results"] if r["check_id"] == "SEC.1")
+        self.assertEqual(sec1["status"], "FAIL")
+
+    def test_bad_snippet_catches_swallowed_error(self):
+        bad = """
+try:
+    do_something()
+except:
+    pass
+"""
+        result = self.server.call_tool("atelier_validate_against_principles", {"snippet": bad})
+        # C1.2 (handles failure cases) should FAIL because of `except: pass`
+        c12 = next(r for r in result["results"] if r["check_id"] == "C1.2")
+        self.assertEqual(c12["status"], "FAIL")
+
+    def test_bad_snippet_catches_obfuscated_names(self):
+        bad = """
+def doStuff(data, temp, x):
+    return data + temp + x
+"""
+        result = self.server.call_tool("atelier_validate_against_principles", {"snippet": bad})
+        # C2.1 (names intent-revealing) should FAIL
+        c21 = next(r for r in result["results"] if r["check_id"] == "C2.1")
+        self.assertEqual(c21["status"], "FAIL")
+
+    def test_result_structure(self):
+        result = self.server.call_tool("atelier_validate_against_principles", {"snippet": "x = 1"})
+        self.assertIn("overall", result)
+        self.assertIn("results", result)
+        self.assertIsInstance(result["results"], list)
+        self.assertGreater(len(result["results"]), 0)
+
+
+if __name__ == "__main__":
+    unittest.main()