feat: vendor upstream Skills and add the GCP provider bundle

Increment 2 of .scratch/solution-architect-agent/spec.md.

- Vendor google-cloud-solution-architecture and the six
  google-cloud-waf-* Skills from google/skills@81a31a6, unmodified,
  with LICENSE, SHA256SUMS and VENDOR.md; add
  scripts/sync-vendored-skills.sh to re-sync at a pinned sha.
- app/agent_skills.py: discover Skills and validate SKILL.md
  frontmatter against the Agent Skills specification.
- app/providers.py + providers/gcp.yaml: load and validate a Cloud
  provider bundle (Skills, output template, per-role models, tool
  tiers, remote MCP servers). MCP allowlists stay empty until the
  read-only review in increment 6.
- Settings CLOUD_PROVIDER and MCP_ENABLE_PREVIEW.
- 35 new tests; full suite 65 passed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
2026-09-15 18:20:17 +01:00
parent 69012b2656
commit 2976dfab21
24 changed files with 3308 additions and 0 deletions

View File

@@ -0,0 +1,90 @@
"""Agent Skills (agentskills.io) discovery and frontmatter validation."""
from pathlib import Path
import pytest
from app.agent_skills import SkillFormatError, discover_skills, read_skill
def write_skill(
parent: Path, dirname: str, frontmatter: str, body: str = "# Instructions\n"
) -> Path:
skill_dir = parent / dirname
skill_dir.mkdir(parents=True)
(skill_dir / "SKILL.md").write_text(
f"---\n{frontmatter}\n---\n\n{body}", encoding="utf-8"
)
return skill_dir
def test_reads_valid_skill_with_optional_fields(tmp_path):
skill_dir = write_skill(
tmp_path,
"pdf-processing",
"name: pdf-processing\n"
"description: >-\n Extract PDF text. Use when handling PDFs.\n"
"license: Apache-2.0\n"
"compatibility: Requires poppler\n"
"metadata:\n category: Documents\n"
"allowed-tools: Read",
)
skill = read_skill(skill_dir)
assert skill.name == "pdf-processing"
assert skill.description == "Extract PDF text. Use when handling PDFs."
assert skill.path == skill_dir
@pytest.mark.parametrize(
("dirname", "frontmatter", "message"),
[
("PDF-Processing", "name: PDF-Processing\ndescription: d", "lowercase"),
("-pdf", "name: -pdf\ndescription: d", "lowercase"),
("pdf-", "name: pdf-\ndescription: d", "lowercase"),
("pdf--processing", "name: pdf--processing\ndescription: d", "lowercase"),
("pdf_processing", "name: pdf_processing\ndescription: d", "lowercase"),
("a" * 65, f"name: {'a' * 65}\ndescription: d", "64"),
("other-dir", "name: pdf-processing\ndescription: d", "directory"),
("pdf", "name: pdf", "description"),
("pdf", "name: pdf\ndescription: ''", "description"),
("pdf", f"name: pdf\ndescription: {'x' * 1025}", "1024"),
("pdf", f"name: pdf\ndescription: d\ncompatibility: {'x' * 501}", "500"),
("pdf", "name: pdf\ndescription: d\nmetadata:\n version: 2", "metadata"),
("pdf", "name: pdf\ndescription: d\nphase: design", "phase"),
],
)
def test_rejects_invalid_frontmatter(tmp_path, dirname, frontmatter, message):
skill_dir = write_skill(tmp_path, dirname, frontmatter)
with pytest.raises(SkillFormatError, match=message):
read_skill(skill_dir)
def test_rejects_skill_without_frontmatter(tmp_path):
skill_dir = tmp_path / "pdf"
skill_dir.mkdir()
(skill_dir / "SKILL.md").write_text("# No frontmatter\n", encoding="utf-8")
with pytest.raises(SkillFormatError, match="frontmatter"):
read_skill(skill_dir)
def test_discovers_immediate_skill_directories_only(tmp_path):
root = tmp_path / "skills"
write_skill(root, "alpha", "name: alpha\ndescription: first")
write_skill(root / "vendor" / "google", "beta", "name: beta\ndescription: nested")
(root / "notes").mkdir()
skills = discover_skills([root, root / "vendor" / "google"])
assert sorted(skills) == ["alpha", "beta"]
def test_rejects_duplicate_skill_names_across_directories(tmp_path):
write_skill(tmp_path / "one", "alpha", "name: alpha\ndescription: first")
write_skill(tmp_path / "two", "alpha", "name: alpha\ndescription: second")
with pytest.raises(SkillFormatError, match="alpha"):
discover_skills([tmp_path / "one", tmp_path / "two"])

163
tests/test_providers.py Normal file
View File

@@ -0,0 +1,163 @@
"""Cloud provider bundle loading and validation."""
from pathlib import Path
import pytest
import yaml
from app.providers import ProviderConfigError, load_provider
REPO_ROOT = Path(__file__).parents[1]
def test_gcp_bundle_loads_vendored_skills_and_template():
bundle = load_provider("gcp")
assert bundle.name == "gcp"
assert "google-cloud-solution-architecture" in bundle.skills
assert {name for name in bundle.skills if name.startswith("google-cloud-waf-")} == {
"google-cloud-waf-security",
"google-cloud-waf-reliability",
"google-cloud-waf-cost-optimization",
"google-cloud-waf-operational-excellence",
"google-cloud-waf-performance-optimization",
"google-cloud-waf-sustainability",
}
assert bundle.output_template == (
REPO_ROOT
/ "skills/vendor/google/google-cloud-solution-architecture/assets/output-template.md"
)
assert bundle.models.default
def test_gcp_bundle_mcp_servers_are_tiered_and_referenced_consistently():
bundle = load_provider("gcp")
general = {
server.id for server in bundle.mcp_servers.values() if server.tier == "general"
}
assert general == {"developer_knowledge"}
assert all(
server.url.startswith("https://") for server in bundle.mcp_servers.values()
)
for role, server_ids in bundle.role_mcp_servers.items():
assert set(server_ids) <= set(bundle.mcp_servers), role
def test_role_mcp_servers_can_exclude_preview_servers():
bundle = load_provider("gcp")
with_preview = {s.id for s in bundle.mcp_servers_for_role("requirements-analyst")}
without_preview = {
s.id
for s in bundle.mcp_servers_for_role(
"requirements-analyst", include_preview=False
)
}
assert "asset_inventory" in with_preview
assert "asset_inventory" not in without_preview
assert bundle.mcp_servers_for_role("no-such-role") == ()
def test_model_for_role_falls_back_to_default():
bundle = load_provider("gcp")
assert bundle.models.for_role("no-such-role") == bundle.models.default
def test_unknown_provider_is_rejected(tmp_path):
(tmp_path / "providers").mkdir()
with pytest.raises(ProviderConfigError, match="aws"):
load_provider("aws", root=tmp_path)
def minimal_bundle(tmp_path: Path, **overrides) -> dict:
skill_dir = tmp_path / "skills" / "alpha"
skill_dir.mkdir(parents=True)
(skill_dir / "SKILL.md").write_text(
"---\nname: alpha\ndescription: d\n---\n", encoding="utf-8"
)
(tmp_path / "template.md").write_text("# Template\n", encoding="utf-8")
config = {
"name": "test",
"skills_dirs": ["skills"],
"output_template": "template.md",
"models": {"default": "gemini/test-model"},
"tools": {"general": ["mermaid_check"], "cloud_read": []},
"mcp_auth": {"kind": "google_adc"},
"mcp_servers": {
"docs": {"url": "https://example.com/mcp", "tier": "general", "allow": []}
},
"role_mcp_servers": {"describer": ["docs"]},
}
config.update(overrides)
return config
def write_bundle(tmp_path: Path, config: dict) -> None:
(tmp_path / "providers").mkdir(exist_ok=True)
(tmp_path / "providers" / "test.yaml").write_text(
yaml.safe_dump(config), encoding="utf-8"
)
def test_minimal_bundle_loads(tmp_path):
write_bundle(tmp_path, minimal_bundle(tmp_path))
bundle = load_provider("test", root=tmp_path)
assert list(bundle.skills) == ["alpha"]
assert bundle.mcp_servers["docs"].allow == ()
assert bundle.mcp_servers["docs"].preview is False
@pytest.mark.parametrize(
("override", "message"),
[
({"name": "other"}, "name"),
({"skills_dirs": ["missing"]}, "skills_dirs"),
({"output_template": "missing.md"}, "output_template"),
({"models": {}}, "models.default"),
({"tools": {"mutating": ["terraform_apply"]}}, "tier"),
(
{
"mcp_servers": {
"docs": {
"url": "http://example.com/mcp",
"tier": "general",
"allow": [],
}
}
},
"https",
),
(
{
"mcp_servers": {
"docs": {
"url": "https://example.com/mcp",
"tier": "write",
"allow": [],
}
}
},
"tier",
),
(
{
"mcp_servers": {
"docs": {"url": "https://example.com/mcp", "tier": "general"}
}
},
"allow",
),
({"role_mcp_servers": {"describer": ["missing"]}}, "missing"),
],
)
def test_invalid_bundles_are_rejected(tmp_path, override, message):
write_bundle(tmp_path, minimal_bundle(tmp_path, **override))
with pytest.raises(ProviderConfigError, match=message):
load_provider("test", root=tmp_path)

View File

@@ -0,0 +1,53 @@
"""Vendored upstream Skills stay byte-for-byte what VENDOR.md says they are."""
import hashlib
from pathlib import Path
from app.agent_skills import discover_skills
VENDOR_DIR = Path(__file__).parents[1] / "skills" / "vendor" / "google"
PINNED_COMMIT = "81a31a6e4300425598a69fd96c8e843c947ae308"
EXPECTED_SKILLS = {
"google-cloud-solution-architecture",
"google-cloud-waf-security",
"google-cloud-waf-reliability",
"google-cloud-waf-cost-optimization",
"google-cloud-waf-operational-excellence",
"google-cloud-waf-performance-optimization",
"google-cloud-waf-sustainability",
}
def read_checksums() -> dict[str, str]:
entries = {}
for line in (VENDOR_DIR / "SHA256SUMS").read_text(encoding="utf-8").splitlines():
digest, relative_path = line.split(maxsplit=1)
entries[relative_path] = digest
return entries
def test_vendor_record_names_pinned_commit_and_licence():
vendor_md = (VENDOR_DIR / "VENDOR.md").read_text(encoding="utf-8")
assert f"`{PINNED_COMMIT}`" in vendor_md
assert "Apache License" in (VENDOR_DIR / "LICENSE").read_text(encoding="utf-8")
def test_checksums_cover_every_vendored_file_and_match():
checksums = read_checksums()
on_disk = {
path.relative_to(VENDOR_DIR).as_posix()
for path in VENDOR_DIR.rglob("*")
if path.is_file() and path.name not in {"SHA256SUMS", "VENDOR.md"}
}
assert set(checksums) == on_disk
for relative_path, digest in checksums.items():
actual = hashlib.sha256((VENDOR_DIR / relative_path).read_bytes()).hexdigest()
assert actual == digest, f"{relative_path} differs from vendored upstream"
def test_expected_upstream_skills_are_present_and_valid():
skills = discover_skills([VENDOR_DIR])
assert set(skills) == EXPECTED_SKILLS