CANP-WP-0006 T04: publish API

POST /packages takes {registry, source, files}, where files is the same shape
GET /archives returns — so an archive round-trips into a publish without
translation and a mirror is a GET followed by a POST. A test verifies the
round-trip by digest rather than asserting it.

Validation is the reference implementation's, applied to the posted files
materialized in a temporary directory. Only reserved paths and
manifest-referenced files are stored (section 2), path traversal is refused,
identical re-publishes are accepted, and different content under the same
id@version is 409 (section 17).

The identity mechanism, stated plainly rather than implied: a single shared
bearer token proving the caller is the operator of this service. It is not
per-publisher identity — every token holder is indistinguishable — and auth.py
says so where someone might otherwise assume more.

With no token configured the service is read-only. That is the correct default
rather than an inconvenience: section 20.1 asks a registry to refuse
publication into a closed namespace it does not consider the publisher to own,
and an unauthenticated service considers nobody to own anything.

Namespace claims live in namespace_claims (migration 0002) and are enforced
here, which a filesystem registry cannot do at all — but only as precisely as
the identity allows. A closed namespace is protected from anonymous callers; it
cannot be attributed among several publishers. Per-publisher identity is
deferred and is the main thing between this and a registry several people can
publish to.

Migration hygiene found while adding 0002: autogenerate proposed an ALTER
COLUMN TYPE on package_files.package_version_id, because the foreign key's type
was left to inference and compared as a variant against a reflected plain type.
SQLite cannot alter a column type, so 0002 failed halfway — table created,
revision unstamped, the partially-applied state that is worst to debug later.
Fixed at the cause: the column is typed explicitly, and 0001 was corrected
rather than patched over, which is legitimate only because it has never run
outside this repo's tests. alembic check now reports no drift.

Health tests now compute the expected migration head from the script directory
instead of hardcoding it, so adding a migration cannot fail them spuriously.

Service tests 22 -> 33; reference unaffected at 99.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Bjefh8NUiEiahN4JLwoSKM

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 388925@bnt-lap001
Assistant-Session: 3507023f-e0fd-4a1e-9d90-a0d4217d1502
This commit is contained in:
tegwick 2026-09-06 20:30:00 +02:00
parent 133e796eff
commit a7e12d5e04
13 changed files with 468 additions and 13 deletions

View file

@ -89,9 +89,42 @@ once the caller says which registry they meant.
Asking for a package without naming a version applies § 17.1's selector rules,
so a prerelease is never chosen implicitly.
## Publish API
`POST /packages` takes `{registry, source, files}`, where `files` is the same
shape `GET /archives` returns — so an archive round-trips into a publish
without translation, and a mirror is a GET followed by a POST.
Validation is the reference implementation's, applied to the posted files
materialized in a temporary directory. Only reserved paths and
manifest-referenced files are stored (§ 2); path traversal is refused.
Re-publishing identical content is accepted, different content under the same
`<id>@<version>` is 409 (§ 17).
### What identity means here
A single shared bearer token (`CANNED_PROMPTS_PUBLISH_TOKEN`), proving the
caller is **the operator of this service** — not per-publisher identity. Every
holder of the token is indistinguishable.
With no token configured the service is **read-only**. That is the correct
default rather than an inconvenience: § 20.1 asks a registry to refuse
publication into a closed namespace it does not consider the publisher to own,
and an unauthenticated service considers nobody to own anything.
Namespace claims (§ 20.1) live in `namespace_claims` and are **enforced** here,
which a filesystem registry cannot do at all — but only as precisely as the
identity behind them. A `closed` namespace is protected from anonymous callers;
it cannot be attributed among several publishers. Until per-publisher identity
exists, a claim's `owner` is documentation rather than an access decision, and
the code says so where it matters.
## Status
`CANP-WP-0006` T01T03 are done: skeleton, health surface, tenant-keyed schema
and migration `0001`, and the read API. The publish API, HTTP registry client
and container image (T04T06) are not built yet. T06 produces the image digest
that `rapp-canned-prompts` needs to pin.
`CANP-WP-0006` T01T04 are done: skeleton, health surface, tenant-keyed schema
(migrations `0001``0002`), and the read and publish APIs. The HTTP registry
client in the CLI and the container image (T05T06) are not built yet. T06
produces the image digest that `rapp-canned-prompts` needs to pin.
Per-publisher identity is deferred, and is the main thing standing between this
and a service that several people can publish to.

View file

@ -1,6 +1,7 @@
[alembic]
script_location = migrations
prepend_sys_path = src
path_separator = os
[loggers]
keys = root

View file

@ -60,7 +60,7 @@ def upgrade() -> None:
op.create_table('package_files',
sa.Column('id', BigIntPK, autoincrement=True, nullable=False),
sa.Column('tenant', sa.String(length=64), nullable=False),
sa.Column('package_version_id', sa.BigInteger(), nullable=False),
sa.Column('package_version_id', BigIntPK, nullable=False),
sa.Column('path', sa.String(length=1024), nullable=False),
sa.Column('content', sa.LargeBinary(), nullable=False),
sa.ForeignKeyConstraint(['package_version_id'], ['package_versions.id'], ondelete='CASCADE'),

View file

@ -0,0 +1,39 @@
"""namespace claims
Registry namespace ownership (section 20.1). Descriptive on a filesystem
registry, which cannot authenticate a publisher; enforced here, which can
though only as strongly as the identity behind it.
Revision ID: 0002
Revises: 0001
"""
from alembic import op
import sqlalchemy as sa
BigIntPK = sa.BigInteger().with_variant(sa.Integer, "sqlite")
revision = '0002'
down_revision = '0001'
branch_labels = None
depends_on = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.create_table('namespace_claims',
sa.Column('id', sa.BigInteger().with_variant(sa.Integer(), 'sqlite'), autoincrement=True, nullable=False),
sa.Column('tenant', sa.String(length=64), nullable=False),
sa.Column('registry', sa.String(length=128), nullable=False),
sa.Column('namespace', sa.String(length=256), nullable=False),
sa.Column('policy', sa.String(length=16), nullable=False),
sa.Column('owner', sa.String(length=256), nullable=True),
sa.PrimaryKeyConstraint('id'),
sa.UniqueConstraint('tenant', 'registry', 'namespace', name='uq_namespace_claim')
)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.drop_table('namespace_claims')
# ### end Alembic commands ###

View file

@ -72,5 +72,5 @@ def create_app(settings: Settings | None = None, engine: Engine | None = None) -
"migration": readiness.migration,
}
app.include_router(make_router(get_session, get_tenant))
app.include_router(make_router(get_session, get_tenant, settings))
return app

View file

@ -0,0 +1,54 @@
"""Publisher identity.
What this is, stated plainly so nobody mistakes it for more: a single shared
bearer token proving the caller is *the operator of this service*. It is not
per-publisher identity every holder of the token is indistinguishable.
That limit decides how § 20.1 namespace claims are enforced here. A `closed`
namespace can be protected from anonymous callers, which a filesystem registry
cannot do at all, but it cannot be attributed to one of several publishers.
Per-publisher identity is deferred; until it exists, a claim's `owner` is
documentation rather than an access decision.
With no token configured the service is read-only. Refusing writes is the
correct default for a registry that cannot tell who is calling § 20.1 asks a
registry to refuse publication into a closed namespace it does not consider the
publisher to own, and an unauthenticated service considers nobody to own
anything.
"""
from __future__ import annotations
import hmac
from fastapi import HTTPException, Request
from .settings import Settings
class PublishDenied(HTTPException):
pass
def require_publisher(request: Request, settings: Settings) -> str:
"""Return the publisher identity, or refuse."""
if not settings.publish_token:
raise PublishDenied(
status_code=503,
detail=(
"publishing is not configured: this service has no publisher "
"identity, so it cannot tell who is calling and refuses writes "
"rather than accepting anonymous publishes (§ 20.1)"
),
)
header = request.headers.get("authorization", "")
scheme, _, credential = header.partition(" ")
if scheme.lower() != "bearer" or not credential:
raise PublishDenied(
status_code=401, detail="publishing requires an Authorization: Bearer token"
)
# compare_digest so a wrong token cannot be recovered by timing the reply.
if not hmac.compare_digest(credential, settings.publish_token):
raise PublishDenied(status_code=403, detail="token not accepted")
return settings.publisher_name

View file

@ -91,8 +91,11 @@ class PackageFile(Base):
id: Mapped[int] = mapped_column(BigIntPK, primary_key=True, autoincrement=True)
tenant: Mapped[str] = mapped_column(String(64), nullable=False)
# Typed explicitly with the same variant as the target key. Left to
# inference, autogenerate compares a variant against a reflected plain type
# and proposes a spurious ALTER COLUMN on every future migration.
package_version_id: Mapped[int] = mapped_column(
ForeignKey("package_versions.id", ondelete="CASCADE"), nullable=False
BigIntPK, ForeignKey("package_versions.id", ondelete="CASCADE"), nullable=False
)
path: Mapped[str] = mapped_column(String(1024), nullable=False)
content: Mapped[bytes] = mapped_column(LargeBinary, nullable=False)
@ -130,3 +133,24 @@ class IndexEntry(Base):
DateTime(timezone=True), nullable=False, default=utcnow
)
last_seen_at: Mapped[dt.datetime | None] = mapped_column(DateTime(timezone=True))
class NamespaceClaim(Base):
"""A registry's namespace ownership claim (§ 20.1).
On a filesystem registry these claims are descriptive, because nothing can
authenticate a publisher. A service *can*, so here they are enforced which
is only as strong as the identity behind them; see `auth.py`.
"""
__tablename__ = "namespace_claims"
__table_args__ = (
UniqueConstraint("tenant", "registry", "namespace", name="uq_namespace_claim"),
)
id: Mapped[int] = mapped_column(BigIntPK, primary_key=True, autoincrement=True)
tenant: Mapped[str] = mapped_column(String(64), nullable=False)
registry: Mapped[str] = mapped_column(String(128), nullable=False)
namespace: Mapped[str] = mapped_column(String(256), nullable=False)
policy: Mapped[str] = mapped_column(String(16), nullable=False, default="open")
owner: Mapped[str | None] = mapped_column(String(256))

View file

@ -10,13 +10,17 @@ swallow a neighbouring segment.
from __future__ import annotations
import base64
import tempfile
from pathlib import Path
from typing import Any
from fastapi import APIRouter, Depends, HTTPException, Query
from fastapi import APIRouter, Body, Depends, HTTPException, Query, Request
from sqlalchemy.orm import Session
from . import store
from .store import Ambiguous, NotFound, Reference, StoreError
from .auth import require_publisher
from .settings import Settings
from .store import Ambiguous, Conflict, NotFound, Reference, StoreError
router = APIRouter()
@ -28,9 +32,31 @@ def as_http(exc: StoreError) -> HTTPException:
# 409, not 300: the request is answerable, but only once the caller says
# which registry they meant.
return HTTPException(status_code=409, detail=str(exc))
if isinstance(exc, Conflict):
return HTTPException(status_code=409, detail=str(exc))
return HTTPException(status_code=400, detail=str(exc))
def decode_files(payload: dict[str, Any]) -> dict[str, bytes]:
"""Accept the same shape `/archives` returns, so an archive round-trips."""
files: dict[str, bytes] = {}
for path, item in (payload or {}).items():
if not isinstance(item, dict):
raise HTTPException(400, f"{path}: expected an object with text or base64")
if "text" in item:
files[path] = str(item["text"]).encode("utf-8")
elif "base64" in item:
try:
files[path] = base64.b64decode(str(item["base64"]), validate=True)
except Exception as exc: # noqa: BLE001
raise HTTPException(400, f"{path}: invalid base64") from exc
else:
raise HTTPException(400, f"{path}: expected text or base64")
if not files:
raise HTTPException(400, "no files supplied")
return files
def version_payload(row: Any, *, manifest: bool = False) -> dict[str, Any]:
payload = {
"registry": row.registry,
@ -50,9 +76,60 @@ def version_payload(row: Any, *, manifest: bool = False) -> dict[str, Any]:
return payload
def make_router(get_session, get_tenant) -> APIRouter:
def make_router(get_session, get_tenant, settings: Settings) -> APIRouter:
api = APIRouter()
@api.post("/packages", status_code=201)
def publish(
request: Request,
payload: dict[str, Any] = Body(...),
session: Session = Depends(get_session),
tenant: str = Depends(get_tenant),
) -> dict[str, Any]:
publisher = require_publisher(request, settings)
registry = str(payload.get("registry") or "").strip()
if not registry:
raise HTTPException(400, "registry is required")
files = decode_files(payload.get("files"))
source = str(payload.get("source") or f"published by {publisher}")
with tempfile.TemporaryDirectory() as tmp:
try:
package_dir = store.materialize(files, Path(tmp))
manifest_id = store.peek_id(package_dir)
except StoreError as exc:
raise as_http(exc) from exc
except Exception as exc: # noqa: BLE001 — malformed package
raise HTTPException(400, f"invalid package: {exc}") from exc
# § 20.1 enforced, not advised: unlike a filesystem registry, a
# service can refuse. It can only refuse as precisely as its
# identity allows — see auth.py.
claim = store.claim_for(session, tenant, registry, manifest_id)
if claim is not None and claim.policy == "closed":
owner = claim.owner or "unspecified"
if publisher != (claim.owner or publisher):
raise HTTPException(
403,
f"namespace {store.namespace_of(manifest_id)!r} in registry "
f"{registry!r} is closed (owner: {owner})",
)
try:
version = store.ingest(
session, tenant, registry, package_dir, source=source, method="publish"
)
session.commit()
except StoreError as exc:
session.rollback()
raise as_http(exc) from exc
except Exception as exc: # noqa: BLE001
session.rollback()
raise HTTPException(400, f"invalid package: {exc}") from exc
return version_payload(version)
@api.get("/packages")
def list_packages(
q: str = Query("", description="Substring over id, name, summary and tags"),

View file

@ -18,6 +18,12 @@ class Settings(BaseSettings):
tenant: str = "default"
service_name: str = "canned-prompts"
# A single shared token, so the identity it proves is "the operator of this
# service" and nothing finer. Unset means the service is read-only; see
# auth.py for why that is the right default rather than an inconvenience.
publish_token: str = ""
publisher_name: str = "operator"
@property
def configured(self) -> bool:
return bool(self.database_url)

View file

@ -248,3 +248,39 @@ def index_entries(session: Session, tenant: str) -> list[IndexEntry]:
session.scalars(select(IndexEntry).where(IndexEntry.tenant == tenant))
)
return sorted(rows, key=lambda r: (r.registry, r.package_id, cp.parse_semver(r.version)))
def namespace_of(package_id: str) -> str:
return package_id.split("/")[0] if "/" in package_id else package_id
def claim_for(session: Session, tenant: str, registry: str, package_id: str):
from .models import NamespaceClaim
return session.scalar(
select(NamespaceClaim).where(
NamespaceClaim.tenant == tenant,
NamespaceClaim.registry == registry,
NamespaceClaim.namespace == namespace_of(package_id),
)
)
def materialize(files: dict[str, bytes], destination: Path) -> Path:
"""Write a posted package to disk so the reference validator can read it."""
for relative, content in files.items():
if relative.startswith("/") or ".." in Path(relative).parts:
raise StoreError(f"unsafe path in package: {relative}")
target = destination / relative
target.parent.mkdir(parents=True, exist_ok=True)
target.write_bytes(content)
return destination
def peek_id(package_dir: Path) -> str:
"""The declared id, needed before validation to check namespace policy."""
manifest = cp.read_manifest(package_dir)
package_id = manifest.get("id")
if not isinstance(package_id, str) or not package_id.strip():
raise StoreError("manifest declares no id")
return package_id

View file

@ -12,6 +12,7 @@ from pathlib import Path
import pytest
from alembic import command
from alembic.config import Config
from alembic.script import ScriptDirectory
from fastapi.testclient import TestClient
from canned_prompts_service.api import create_app
@ -21,6 +22,13 @@ from canned_prompts_service.settings import Settings
ROOT = Path(__file__).resolve().parents[1]
def expected_head() -> str:
"""Computed, not hardcoded: a new migration must not fail these tests."""
config = Config(str(ROOT / "alembic.ini"))
config.set_main_option("script_location", str(ROOT / "migrations"))
return ScriptDirectory.from_config(config).get_current_head()
def migrated_url(tmp_path: Path) -> str:
url = f"sqlite:///{tmp_path / 'svc.db'}"
config = Config(str(ROOT / "alembic.ini"))
@ -70,7 +78,7 @@ def test_readyz_fails_when_the_schema_is_not_migrated(tmp_path: Path) -> None:
def test_readyz_reports_the_migration_when_ready(ready_client: TestClient) -> None:
body = ready_client.get("/readyz").json()
assert body["ready"] is True
assert body["migration"] == "0001"
assert body["migration"] == expected_head()
def test_state_health_matches_the_fleet_shape(ready_client: TestClient) -> None:
@ -78,7 +86,7 @@ def test_state_health_matches_the_fleet_shape(ready_client: TestClient) -> None:
assert body["status"] == "ok"
assert body["service"] == "canned-prompts"
assert body["db"] == "connected"
assert body["migration"] == "0001"
assert body["migration"] == expected_head()
def test_state_health_degrades_rather_than_lying() -> None:

View file

@ -0,0 +1,145 @@
"""Publish API (CANP-WP-0006 T04)."""
from __future__ import annotations
from pathlib import Path
import pytest
from fastapi.testclient import TestClient
from sqlalchemy.orm import Session
from canned_prompts_service.api import create_app
from canned_prompts_service.db import make_engine, make_session_factory
from canned_prompts_service.models import NamespaceClaim
from canned_prompts_service.settings import Settings
REPO = Path(__file__).resolve().parents[2]
TOKEN = "test-token"
AUTH = {"Authorization": f"Bearer {TOKEN}"}
def package_payload(name: str, registry: str = "local") -> dict:
"""Built from the archive shape, so an archive round-trips into a publish."""
source_dir = REPO / "examples" / name
import canned_prompts as cp
manifest = cp.validate_package(source_dir)
files = {
rel: {"text": (source_dir / rel).read_text(encoding="utf-8")}
for rel in sorted(cp.package_members(source_dir, manifest))
}
return {"registry": registry, "source": f"examples/{name}", "files": files}
@pytest.fixture()
def writable(db_url: str) -> TestClient:
settings = Settings(database_url=db_url, publish_token=TOKEN)
return TestClient(create_app(settings, make_engine(db_url)))
def test_unconfigured_service_refuses_writes(client: TestClient) -> None:
"""No publisher identity means the service cannot tell who is calling."""
response = client.post("/packages", json=package_payload("house-style"))
assert response.status_code == 503
assert "publishing is not configured" in response.json()["detail"]
def test_publish_requires_a_token(writable: TestClient) -> None:
response = writable.post("/packages", json=package_payload("house-style"))
assert response.status_code == 401
def test_wrong_token_is_refused(writable: TestClient) -> None:
response = writable.post(
"/packages",
json=package_payload("house-style"),
headers={"Authorization": "Bearer nope"},
)
assert response.status_code == 403
def test_publish_then_read_back(writable: TestClient) -> None:
response = writable.post("/packages", json=package_payload("house-style"), headers=AUTH)
assert response.status_code == 201
assert response.json()["reference"] == "local:practice/house-style@0.1.1"
body = writable.get("/packages/local:practice/house-style@0.1.1").json()
assert body["type"] == "fragment"
assert writable.get("/index").json()["count"] == 1
def test_archive_round_trips_into_a_publish(writable: TestClient) -> None:
"""What GET /archives returns is exactly what POST /packages accepts."""
writable.post("/packages", json=package_payload("house-style"), headers=AUTH)
archive = writable.get("/archives/practice/house-style@0.1.1").json()
republished = writable.post(
"/packages",
json={"registry": "mirror", "source": "round-trip", "files": archive["files"]},
headers=AUTH,
)
assert republished.status_code == 201
assert republished.json()["content_digest"] == archive["content_digest"]
def test_identical_republish_is_accepted(writable: TestClient) -> None:
payload = package_payload("house-style")
assert writable.post("/packages", json=payload, headers=AUTH).status_code == 201
assert writable.post("/packages", json=payload, headers=AUTH).status_code == 201
def test_different_content_under_the_same_version_is_a_conflict(writable: TestClient) -> None:
"""§ 17: a published id@version is immutable within a registry."""
payload = package_payload("house-style")
writable.post("/packages", json=payload, headers=AUTH)
payload["files"]["prompt.md"]["text"] += "\nsomething different\n"
response = writable.post("/packages", json=payload, headers=AUTH)
assert response.status_code == 409
assert "already exists with different content" in response.json()["detail"]
def test_invalid_package_is_rejected(writable: TestClient) -> None:
payload = package_payload("house-style")
payload["files"]["prompt.yaml"]["text"] = "format: canned-prompt/v9.9\nid: x/y\n"
response = writable.post("/packages", json=payload, headers=AUTH)
assert response.status_code == 400
def test_path_traversal_is_refused(writable: TestClient) -> None:
payload = package_payload("house-style")
payload["files"]["../escape.txt"] = {"text": "no"}
assert writable.post("/packages", json=payload, headers=AUTH).status_code == 400
def test_closed_namespace_refuses_a_publisher_it_cannot_attribute(
db_url: str, session: Session
) -> None:
"""§ 20.1 enforced rather than advised — as far as identity allows."""
session.add(
NamespaceClaim(
tenant="default", registry="local", namespace="practice",
policy="closed", owner="someone-else",
)
)
session.commit()
client = TestClient(
create_app(Settings(database_url=db_url, publish_token=TOKEN), make_engine(db_url))
)
response = client.post("/packages", json=package_payload("house-style"), headers=AUTH)
assert response.status_code == 403
assert "is closed" in response.json()["detail"]
def test_open_namespace_publishes(db_url: str, session: Session) -> None:
session.add(
NamespaceClaim(
tenant="default", registry="local", namespace="practice", policy="open"
)
)
session.commit()
client = TestClient(
create_app(Settings(database_url=db_url, publish_token=TOKEN), make_engine(db_url))
)
assert client.post(
"/packages", json=package_payload("house-style"), headers=AUTH
).status_code == 201

View file

@ -155,7 +155,7 @@ models and the same migration.
```task
id: CANP-WP-0006-T04
status: todo
status: done
priority: high
state_hub_task_id: "929932b6-931b-5409-a648-b6814f07f0b4"
```
@ -170,6 +170,38 @@ filesystem registry a service *can* authenticate a publisher. Note what identity
mechanism it uses; if there is none yet, say so and refuse writes rather than
accepting anonymous publishes into a closed namespace.
**Done.** `POST /packages` takes the same `files` shape `GET /archives`
returns, so an archive round-trips into a publish without translation and a
mirror is a GET followed by a POST — verified by a test, not just asserted.
**The identity mechanism, stated plainly.** A single shared bearer token
proving the caller is *the operator of this service*. Not per-publisher
identity: every token holder is indistinguishable. With no token configured the
service is read-only, which is the right default rather than an inconvenience —
§ 20.1 asks a registry to refuse publication into a closed namespace it does not
consider the publisher to own, and an unauthenticated service considers nobody
to own anything.
Namespace claims live in `namespace_claims` (migration `0002`) and are enforced,
which a filesystem registry cannot do at all — but only as precisely as the
identity allows. A closed namespace is protected from anonymous callers; it
cannot be attributed among several publishers. Until per-publisher identity
exists, a claim's `owner` is documentation rather than an access decision, and
`auth.py` says so rather than letting the code imply more than it delivers.
**Handed forward:** per-publisher identity is the main thing between this and a
registry several people can publish to.
**Migration hygiene found while adding `0002`.** Autogenerate proposed an
`ALTER COLUMN TYPE` on `package_files.package_version_id`, because the foreign
key's type was left to inference and compared as a variant against a reflected
plain type. SQLite cannot alter a column type, so `0002` failed halfway,
leaving the table created and the revision unstamped — the partially-applied
state that is worst to debug later. Fixed at the cause: the column is typed
explicitly, and `0001` was corrected rather than patched over, which is
legitimate only because it has never run outside this repo's tests. `alembic
check` now reports no drift.
## HTTP registry in the reference CLI
```task