agnes-the-ai-analyst/tests/test_store_entity_versions.py
Vojtech a694a30a5e
fix(store): surface review failures + harden publish gate (#316)
* fix(store): surface review failures + harden publish gate

Four independent fixes to the flea-market submission pipeline, all surfaced
by an admin upload that landed at status='approved' without an LLM review.

1. LLM truncation no longer pins submissions in review_error.
   - Raised MAX_RESPONSE_TOKENS 2500 → 6000 in llm_review.py
   - Added one-shot retry-with-doubled-budget in anthropic_provider.py
     (capped at 4× initial)

2. Flea detail page surfaces the latest submission's failure verdict even
   when a previously-approved version is still serving (deferred-promotion
   path). The _quarantine_banner gate widened from `visibility != approved`
   to also fire on `blocked_inline / blocked_llm / review_error`, with copy
   that distinguishes the v2+ edit case ("Latest edit failed review —
   previously approved version (vN) keeps serving") from the initial-upload
   quarantine wording.

3. Restore button + endpoint no longer allow restoring a version that was
   never approved. Added StoreEntitiesRepository.get_with_version_approvals
   joining store_submissions, gated the UI button on submission_status in
   ('approved', None), rendered status pills for non-restorable rows, and
   added a 400 version_not_approved guard in POST /restore.

4. **BREAKING (operator-facing)**: publish gate is now fail-CLOSED on
   misconfig. The previous get_guardrails_enabled() silently fell back to
   "disabled, auto-approve everything" when guardrails.enabled=true in YAML
   but no ANTHROPIC_API_KEY was in env. Split into:
     - get_guardrails_enabled()              (intent — YAML)
     - get_guardrails_llm_provider_ready()   (readiness — env)
   Three-state matrix:
     enabled=false                       → auto-approve (unchanged)
     enabled=true + ready=true           → normal pipeline (unchanged)
     enabled=true + ready=false (NEW)    → submissions hold at pending_llm
                                           awaiting admin retry or override
                                           (was: silent auto-approve)
   Admin "Retry review" eligibility broadened to include pending_llm.
   Boot-time WARNING banner surfaces the misconfig in app/main.py.
   docs/STORE_GUARDRAILS.md updated with the three-state matrix.
   Operators relying on the auto-fallback for local-dev no-LLM setups must
   now explicitly set `guardrails.enabled: false` in instance.yaml.

Tests: 4623 passed. Added TestPublishGateFailClosed (4 tests) and
TestRestoreVersion::test_restore_rejects_* (3 tests). conftest.py adds an
autouse fixture defaulting guardrails OFF so legacy tests don't need to
know about the new toggle.

* fix(store): admin override promotes v2+ edits to current

The override handler at app/api/admin.py:3708 only flipped submission
status → 'overridden' and entity visibility → 'approved'. Under the v37+
deferred-promotion model that's insufficient for v2+ edits / restores:
the new bundle sits in versions/v<N>/plugin/ and the entity row stays at
the prior approved version_no + hash + on-disk live bundle. Installers
kept getting the OLD bytes the admin had just intended to replace.

Mirror the runner.run_llm_review auto-approval branch: look up the
submission's version_hash in entity.version_history, and if its `n`
differs from entity.version_no, promote_version + _swap_live_to_version.
Initial v1 overrides are unaffected — the loop finds n=1 == version_no
and skips promotion.

Tests:
- test_override_v2_edit_promotes_to_current: stage v1 approved + v2
  blocked_llm; override the v2 sub; assert entity.version_no=2,
  entity.version flips off the v1 hash, and the live plugin/ dir
  mirrors versions/v2/plugin/.
- test_override_v1_initial_upload_no_promote: regression guard so the
  promote loop doesn't accidentally bump a v1 override.

Audit log gains a promoted_to_version_no field on the override action.

* fix(store): retry/rescan review staged bundle; override forward-only

Two adversarial-review findings from a Codex pass on the publish-gate
work.

C1. Admin retry + rescan were passing live `plugin/` to the LLM. For a
v2+ submission held at `pending_llm` / `blocked_llm` / `review_error`,
live still holds the prior approved version's bytes — so the LLM
reviewed the WRONG bytes, and the runner's hash-match promotion in
`run_llm_review` would then advance the entity to staged bytes that
were never actually reviewed. Resolve the staged
`<entity>/versions/v<N>/plugin/` from the submission's
`version_history` entry, with a fall-back to live for legacy pre-v37
rows that never seeded a versions/ dir. Helpers
`_submission_plugin_dir` and `_version_no_for_submission` added to
`app/api/store.py` so override / retry / rescan share one path.

H1. Override's promote loop used `target != current`, which would
silently demote the live bundle when admin overrode a stale v2
submission while v3 was already approved + live. Changed to
`target > current` so override flips status + visibility on the row
regardless, but on-disk promotion only fires forward. Same `>`
defensive guard applied in `runner.run_llm_review` so a late LLM
verdict racing with a newer approval can't demote either.

Tests:
- TestAdminRetryReviewsStagedBundle::test_retry_v2_blocked_passes_staged_dir_not_live
- TestAdminRetryReviewsStagedBundle::test_rescan_v2_blocked_passes_staged_dir_not_live
- TestOverrideForwardOnly::test_override_stale_v2_does_not_demote_when_v3_current

* review polish: CHANGELOG drift, override eligibility, defensive copy

Three small additions on top of the retry/rescan staged-bundle fix:

1. CHANGELOG: the PR's bullets had drifted into the released
   [0.54.17] section during rebase (context-match landed them next
   to already-released content). Moved them up to [Unreleased] where
   they belong; [0.54.17] now holds only what was actually released
   (refresh-marketplace ls-remote, /me/activity hero, CI sharding +
   workflow polish).

2. app/api/admin.py: admin override eligibility now accepts
   pending_llm alongside blocked_inline + blocked_llm + review_error.
   Closes a UX gap from the new fail-CLOSED behavior: under
   enabled-but-not-ready, a known-good submission would otherwise
   sit indefinitely until the admin set credentials AND clicked
   Retry. Override already routes through version_history (and is
   now forward-only on promote), so it stays safe for v2+ deferred-
   promotion submissions.

3. src/repositories/store_entities.py: get_with_version_approvals
   defensively copies each version_history entry before annotating
   with submission_status. self.get() re-parses JSON each call today
   so this is belt-and-suspenders against any future caching layer
   leaking the annotated key into a subsequent plain get() call.

Tests: 112 passed (focused on test_store_entity_versions +
test_admin_store_submissions, covering the retry/rescan staged-
bundle fix the author shipped + this polish).

---------

Co-authored-by: ZdenekSrotyr <zdenek.srotyr@keboola.com>
2026-05-15 15:52:07 +02:00

1410 lines
59 KiB
Python

"""v37 flea-market edit feature with version history.
Covers:
* Bundle update bumps version_no + appends version_history entry.
* Metadata-only edit doesn't bump version.
* Type change rejected with 400 type_locked.
* Block-while-pending: 409 prior_version_pending.
* Display name change renames the on-disk slug for live + version dirs.
* Restore copies a prior version forward as v<max+1>; live + history
reflect the new version; original version row keeps its own verdict.
* Restore re-runs guardrails (blocked path leaves live untouched).
* Versions card on detail page renders for owner/admin only.
"""
from __future__ import annotations
import io
import json
import zipfile
from pathlib import Path
import pytest
from argon2 import PasswordHasher
from fastapi.testclient import TestClient
from app.utils import get_store_dir
from src.db import close_system_db, get_system_db
from src.repositories.store_entities import StoreEntitiesRepository
from src.repositories.users import UserRepository
# Strong default description that clears the content guardrail's
# per-component bar (30 chars + 4 distinct words, no placeholder
# leftovers). Tests don't assert on its contents — they just need a
# value that passes review so we can exercise the edit/version path.
_OK_DESC = "Use when validating store version edit flow across every guardrail tier"
@pytest.fixture
def web_client(tmp_path, monkeypatch):
monkeypatch.setenv("DATA_DIR", str(tmp_path))
monkeypatch.setenv("TESTING", "1")
monkeypatch.setenv("JWT_SECRET_KEY", "test-secret-key-min-32-characters!!")
(tmp_path / "state").mkdir()
(tmp_path / "analytics").mkdir()
(tmp_path / "extracts").mkdir()
close_system_db()
from app.main import create_app
app = create_app()
yield TestClient(app)
close_system_db()
def _create_user(client, email, password="UserPass1!"):
ph = PasswordHasher()
conn = get_system_db()
user_id = email.split("@")[0]
UserRepository(conn).create(
id=user_id, email=email, name=user_id, password_hash=ph.hash(password),
)
conn.close()
r = client.post("/auth/token", json={"email": email, "password": password})
assert r.status_code == 200, r.text
return user_id, {"access_token": r.json()["access_token"]}
def _create_admin(client, email="admin-edit@x.com"):
from tests.helpers.auth import grant_admin
user_id, cookies = _create_user(client, email, password="AdminPass1!")
conn = get_system_db()
grant_admin(conn, user_id)
conn.close()
return user_id, cookies
def _make_skill_zip(skill_name: str, body: str = "Body line explaining the skill. " * 12) -> bytes:
buf = io.BytesIO()
with zipfile.ZipFile(buf, "w") as zf:
zf.writestr(
f"{skill_name}/SKILL.md",
f"---\nname: {skill_name}\ndescription: Use when verifying clean-bundle edits across the version-history lifecycle\n---\n\n"
+ body,
)
return buf.getvalue()
def _make_eval_skill_zip(skill_name: str) -> bytes:
buf = io.BytesIO()
with zipfile.ZipFile(buf, "w") as zf:
zf.writestr(
f"{skill_name}/SKILL.md",
f"---\nname: {skill_name}\ndescription: Use when verifying static-security rejects eval-using upload bundles cleanly\n---\n\n"
+ ("Body line explaining the skill. " * 12),
)
zf.writestr(f"{skill_name}/run.sh", "#!/bin/sh\neval $1\n")
return buf.getvalue()
def _upload_clean(client, cookies, name="ed1"):
r = client.post(
"/api/store/entities",
files={"file": ("s.zip", _make_skill_zip(name), "application/zip")},
data={"type": "skill", "description": _OK_DESC}, cookies=cookies,
)
assert r.status_code == 201, r.text
return r.json()["id"]
class TestEditFeature:
def test_metadata_only_edit_no_version_bump(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "metaowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="metaedit")
r = web_client.put(
f"/api/store/entities/{eid}",
data={"description": "Updated description text"},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
entity = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert entity["description"] == "Updated description text"
assert entity["version_no"] == 1
assert len(entity["version_history"]) == 1
def test_bundle_edit_bumps_version_and_appends_history(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "bundleowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="bundleedit")
# PUT with new bundle bytes.
new_zip = _make_skill_zip("bundleedit", body="V2 body. " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", new_zip, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
entity = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert entity["version_no"] == 2
assert len(entity["version_history"]) == 2
v1, v2 = entity["version_history"]
assert v1["n"] == 1
assert v2["n"] == 2
assert v2["hash"] != v1["hash"], "v2 hash must differ from v1"
def test_type_change_rejected(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "typeowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="typelock")
r = web_client.put(
f"/api/store/entities/{eid}",
data={"type": "agent", "description": _OK_DESC},
cookies=owner_cookies,
)
assert r.status_code == 400, r.text
assert r.json()["detail"]["code"] == "type_locked"
def test_block_while_prior_pending_409(self, web_client):
"""Manually flip the entity to visibility=pending + create a
pending submission, then attempt edit → 409."""
from src.repositories.store_submissions import StoreSubmissionsRepository
owner_id, owner_cookies = _create_user(web_client, "blockowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="blockpending")
conn = get_system_db()
# Force pending state.
conn.execute(
"UPDATE store_entities SET visibility_status = 'pending' WHERE id = ?",
[eid],
)
StoreSubmissionsRepository(conn).create(
submitter_id=owner_id, submitter_email="blockowner@x.com",
type="skill", name="blockpending", version="2.0.0",
status="pending_llm", entity_id=eid,
inline_checks={"manifest": {"status": "pass"}},
)
conn.close()
r = web_client.put(
f"/api/store/entities/{eid}",
data={"description": "Trying to edit"},
cookies=owner_cookies,
)
assert r.status_code == 409, r.text
assert r.json()["detail"]["code"] == "prior_version_pending"
def test_name_change_renames_baked_slug(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "renameowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="oldname")
r = web_client.put(
f"/api/store/entities/{eid}",
data={"name": "newname"},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
entity = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert entity["name"] == "newname"
plugin_dir = Path(get_store_dir()) / eid / "plugin"
new_skill_dir = plugin_dir / "skills" / "newname-by-renameowner"
old_skill_dir = plugin_dir / "skills" / "oldname-by-renameowner"
assert new_skill_dir.is_dir(), "renamed slug missing on disk"
assert not old_skill_dir.exists(), "old slug must be gone"
class TestRestoreVersion:
def test_restore_creates_new_version_with_old_bundle(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "restoreowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="restoreme")
# Edit to v2.
v2_zip = _make_skill_zip("restoreme", body="VERSION-2-BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2_zip, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Capture v1 + v2 hashes from history.
conn = get_system_db()
entity = StoreEntitiesRepository(conn).get(eid)
conn.close()
v1_hash = entity["version_history"][0]["hash"]
v2_hash = entity["version_history"][1]["hash"]
assert v1_hash != v2_hash
# Restore v1 → creates v3 with v1's bundle hash.
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
entity = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert entity["version_no"] == 3
assert len(entity["version_history"]) == 3
v3 = entity["version_history"][2]
assert v3["n"] == 3
assert v3["hash"] == v1_hash, (
"restored bundle should hash identically to v1 — same bytes"
)
def test_restore_already_current_400(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "alreadyowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="already")
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=owner_cookies,
)
assert r.status_code == 400, r.text
assert r.json()["detail"]["code"] == "already_current"
def test_restore_unknown_version_404(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "unknownver@x.com")
eid = _upload_clean(web_client, owner_cookies, name="ukver")
r = web_client.post(
f"/api/store/entities/{eid}/versions/99/restore",
cookies=owner_cookies,
)
assert r.status_code == 404, r.text
assert r.json()["detail"]["code"] == "version_not_found"
def test_non_owner_non_admin_cannot_restore(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "owrestore@x.com")
eid = _upload_clean(web_client, owner_cookies, name="ownedver")
v2 = _make_skill_zip("ownedver", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
_, snoop_cookies = _create_user(web_client, "snoopver@x.com")
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=snoop_cookies,
)
assert r.status_code in (403, 404), r.text
def test_restore_rejects_blocked_llm_version(self, web_client, monkeypatch):
"""A v2 that LLM-blocked sits in version_history with
submission.status='blocked_llm'. The restore endpoint must
refuse to roll forward from that bundle — defense in depth
against the UI being bypassed by direct POST."""
# Mock LLM to BLOCK v2.
def mock_review_bundle(*args, **kwargs):
return {
"risk_level": "high",
"summary": "mock block",
"findings": [{"severity": "high", "category": "test",
"file": "x", "explanation": "mock"}],
"template_placeholders_found": 0,
"reviewed_by_model": "mock-model",
"error": None,
}
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
mock_review_bundle,
)
owner_id, owner_cookies = _create_user(web_client, "blockrestore@x.com")
# Phase 1: guardrails OFF — v1 lands approved.
eid = _upload_clean(web_client, owner_cookies, name="blockrestore")
# Phase 2: guardrails ON → v2 blocked, entity stays at v1.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
v2 = _make_skill_zip("blockrestore", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Drive the BG review synchronously.
from src.repositories.store_submissions import StoreSubmissionsRepository
from src.store_guardrails.runner import run_llm_review
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
# Now POST a restore /versions/2/restore. Must 400 because v2
# was never approved.
r = web_client.post(
f"/api/store/entities/{eid}/versions/2/restore",
cookies=owner_cookies,
)
assert r.status_code == 400, r.text
body = r.json()
assert body["detail"]["code"] == "version_not_approved"
assert body["detail"]["source_status"] == "blocked_llm"
def test_restore_rejects_review_error_version(self, web_client, monkeypatch):
"""Same as blocked_llm but the LLM call errored — the
submission row lands at 'review_error' and the version is
equally not-approvable."""
def mock_review_bundle(*args, **kwargs):
return {
"risk_level": None,
"summary": None,
"findings": [],
"template_placeholders_found": 0,
"reviewed_by_model": "mock-model",
"error": "LLMFormatError: mock truncation",
}
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
mock_review_bundle,
)
owner_id, owner_cookies = _create_user(web_client, "errrestore@x.com")
eid = _upload_clean(web_client, owner_cookies, name="errrestore")
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
v2 = _make_skill_zip("errrestore", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
from src.repositories.store_submissions import StoreSubmissionsRepository
from src.store_guardrails.runner import run_llm_review
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
r = web_client.post(
f"/api/store/entities/{eid}/versions/2/restore",
cookies=owner_cookies,
)
assert r.status_code == 400, r.text
body = r.json()
assert body["detail"]["code"] == "version_not_approved"
assert body["detail"]["source_status"] == "review_error"
def test_restore_allows_legacy_v1_without_submission_id(self, web_client):
"""The v1 seed entry created by ``StoreEntitiesRepository.create``
carries ``submission_id=None`` until the API layer backfills.
A restore targeting v1 must NOT be rejected just because the
join can't find a submission status — back-compat for entities
created before v37."""
owner_id, owner_cookies = _create_user(web_client, "legv1@x.com")
eid = _upload_clean(web_client, owner_cookies, name="legv1")
# Manually clear v1's submission_id to simulate the legacy seed.
conn = get_system_db()
repo = StoreEntitiesRepository(conn)
ent = repo.get(eid)
history = ent["version_history"]
history[0]["submission_id"] = None
conn.execute(
"UPDATE store_entities SET version_history = ? WHERE id = ?",
[json.dumps(history), eid],
)
conn.close()
# PUT a v2 so v1 is no longer current.
v2 = _make_skill_zip("legv1", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Restore v1 → must succeed (200), because legacy v1 has
# submission_id=None which the guard treats as approved.
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
class TestEditPage:
def test_edit_page_renders_for_owner(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "editpage@x.com")
eid = _upload_clean(web_client, owner_cookies, name="editrender")
r = web_client.get(
f"/marketplace/flea/{eid}/edit", cookies=owner_cookies,
)
assert r.status_code == 200
assert "edit-form" in r.text
assert "editrender" in r.text
def test_edit_page_404_for_non_owner(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "owneredit@x.com")
eid = _upload_clean(web_client, owner_cookies, name="snoopedit")
_, snoop_cookies = _create_user(web_client, "snoopedit@x.com")
r = web_client.get(
f"/marketplace/flea/{eid}/edit", cookies=snoop_cookies,
)
assert r.status_code == 404
def test_versions_card_renders_for_owner(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "vowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="vcard")
r = web_client.get(
f"/marketplace/flea/{eid}", cookies=owner_cookies,
)
assert r.status_code == 200
assert "versions-card" in r.text
assert "Versions (1)" in r.text
def test_versions_card_hidden_for_non_owner(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "vowner2@x.com")
eid = _upload_clean(web_client, owner_cookies, name="vhide")
_, other_cookies = _create_user(web_client, "vother@x.com")
r = web_client.get(
f"/marketplace/flea/{eid}", cookies=other_cookies,
)
assert r.status_code == 200
assert "versions-card" not in r.text
class TestInstallerAlwaysGetsLatestApproved:
"""Critical contract: existing installers continue receiving the
last APPROVED version through the review window of a new edit, and
NEVER receive an unapproved version. If the new version is blocked,
they keep the prior approved one. If approved, they advance.
Implemented via deferred promotion: PUT/restore append the new
version to history at status='pending_llm' but DO NOT swap live
or bump entity.version_no. runner.run_llm_review's approval branch
promotes; on block, nothing changes.
"""
def _install_as_user(self, web_client, owner_cookies, eid):
"""Install as a separate consumer user, return their list_for_user
rows post-install (mirrors what marketplace.zip serves)."""
from src.repositories.user_store_installs import UserStoreInstallsRepository
installer_id, installer_cookies = _create_user(web_client, "installer@x.com")
r = web_client.post(
f"/api/store/entities/{eid}/install",
cookies=installer_cookies,
)
assert r.status_code == 200, r.text
return installer_id, installer_cookies
def test_pending_review_does_not_break_existing_installer(self, web_client, monkeypatch):
"""Initial upload runs with guardrails OFF (lands approved).
Then we flip guardrails ON and PUT a new bundle. The new
version should defer promotion: existing installer must
continue seeing v1 + entity.version_no=1, not get hidden by a
flipped visibility or a half-promoted live dir."""
from app import instance_config as ic
# Stub LLM scheduling so the BG path never actually runs.
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
owner_id, owner_cookies = _create_user(web_client, "stickyowner@x.com")
# Phase 1: guardrails OFF → initial upload lands approved.
eid = _upload_clean(web_client, owner_cookies, name="sticky")
# Capture v1 hash + size baseline.
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
v1_hash = ent["version"]
v1_size = ent["file_size"]
conn.close()
# Install as a different user.
installer_id, _ = self._install_as_user(web_client, owner_cookies, eid)
# Phase 2: flip guardrails ON for the PUT call. Now an edit
# defers promotion until LLM approves.
# Patch where update_entity looks it up — `from app.instance_config
# import get_guardrails_enabled` binds the symbol into app.api.store,
# so patching the source module isn't enough.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
# PUT a new bundle. Guardrails on → submission lands at
# pending_llm; promotion deferred.
v2_zip = _make_skill_zip("sticky", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2_zip, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Installer's list_for_user must STILL see entity with v1 hash
# + size + visibility='approved'. The pending edit must not
# have hidden the entity from them.
from src.repositories.user_store_installs import UserStoreInstallsRepository
conn = get_system_db()
installs = UserStoreInstallsRepository(conn).list_for_user(installer_id)
ids = {r["id"] for r in installs}
assert eid in ids, (
"installer lost access to the entity during the LLM "
"review window — list_for_user filter excluded it"
)
row = next(r for r in installs if r["id"] == eid)
assert row["version"] == v1_hash, (
f"installer should still get v1 hash {v1_hash[:8]} but "
f"got {row['version'][:8]}"
)
assert row["file_size"] == v1_size, "size shouldn't change pre-promotion"
assert row["visibility_status"] == "approved", (
"entity must stay 'approved' through the review window so "
"existing installers continue serving"
)
# Entity row's version_no must NOT have bumped yet.
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert ent["version_no"] == 1, (
f"version_no must stay 1 during pending review; got {ent['version_no']}"
)
# But version_history MUST have v2 entry tracked (with the
# new hash) so admin can see what's in flight.
history_n = [int(e["n"]) for e in ent["version_history"]]
assert 2 in history_n, "v2 entry must be in history despite no promotion"
def test_blocked_new_version_keeps_installer_on_prior(self, web_client, monkeypatch):
"""Mock the LLM to BLOCK the v2 review. Installer must keep
v1; entity.version_no must stay at 1; live plugin/ must hold
v1's bytes."""
from app import instance_config as ic
# Mock the runner's LLM call to return a high-risk verdict.
def mock_review_bundle(*args, **kwargs):
return {
"risk_level": "high",
"summary": "mock block",
"findings": [{"severity": "high", "category": "test",
"file": "x", "explanation": "mock"}],
"template_placeholders_found": 0,
"reviewed_by_model": "mock-model",
"error": None,
}
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
mock_review_bundle,
)
# Phase 1: initial upload guardrails OFF.
owner_id, owner_cookies = _create_user(web_client, "blockowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="blocksticky")
# Phase 2: switch guardrails ON before PUT.
# Patch where update_entity looks it up — `from app.instance_config
# import get_guardrails_enabled` binds the symbol into app.api.store,
# so patching the source module isn't enough.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
conn = get_system_db()
v1_hash = StoreEntitiesRepository(conn).get(eid)["version"]
conn.close()
installer_id, _ = self._install_as_user(web_client, owner_cookies, eid)
# Edit. Inline checks pass; LLM mocked to block.
v2_zip = _make_skill_zip("blocksticky", body="v2-content " * 80)
# Run the LLM synchronously by calling runner directly after
# the PUT (the BG task may not have fired in TestClient).
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2_zip, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Find the just-created submission + run runner against it.
from src.repositories.store_submissions import StoreSubmissionsRepository
from src.store_guardrails.runner import run_llm_review
from app.utils import get_store_dir
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
# Installer must STILL see v1.
from src.repositories.user_store_installs import UserStoreInstallsRepository
conn = get_system_db()
installs = UserStoreInstallsRepository(conn).list_for_user(installer_id)
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
row = next(r for r in installs if r["id"] == eid)
assert row["version"] == v1_hash, (
f"after v2 blocked, installer must still get v1 hash; got {row['version'][:8]}"
)
assert ent["version_no"] == 1, (
f"version_no must stay at 1 after a blocked verdict; got {ent['version_no']}"
)
def test_approved_new_version_promotes_to_installer(self, web_client):
"""Default test path: guardrails OFF → guardrails-disabled
promote-inline branch fires immediately. Installer's next
list_for_user reflects v2."""
owner_id, owner_cookies = _create_user(web_client, "promoowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="promosticky")
conn = get_system_db()
v1_hash = StoreEntitiesRepository(conn).get(eid)["version"]
conn.close()
installer_id, _ = self._install_as_user(web_client, owner_cookies, eid)
v2_zip = _make_skill_zip("promosticky", body="promo-v2 " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2_zip, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Installer should now see v2 hash.
from src.repositories.user_store_installs import UserStoreInstallsRepository
conn = get_system_db()
installs = UserStoreInstallsRepository(conn).list_for_user(installer_id)
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert ent["version_no"] == 2
row = next(r for r in installs if r["id"] == eid)
assert row["version"] != v1_hash, "installer should advance to v2"
class TestAdminAccess:
"""Admin can edit + restore on entities they don't own (parity with
the existing admin override path)."""
def test_admin_can_edit_non_owned_entity_metadata(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "adminedit-owner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="adminedit")
_, admin_cookies = _create_admin(web_client)
r = web_client.put(
f"/api/store/entities/{eid}",
data={"description": "moderated by admin"},
cookies=admin_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert ent["description"] == "moderated by admin"
assert ent["version_no"] == 1
def test_admin_can_restore_non_owned_entity(self, web_client):
owner_id, owner_cookies = _create_user(web_client, "adminrestore-owner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="adminrestore")
v2 = _make_skill_zip("adminrestore", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
_, admin_cookies = _create_admin(web_client)
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=admin_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert ent["version_no"] == 3
class TestRestoreDeferredPromotion:
"""Restore endpoint mirrors PUT semantics: live + version_no stay
on prior current until LLM approves the restored copy."""
def test_restore_with_guardrails_on_does_not_promote_until_approved(
self, web_client, monkeypatch,
):
"""Owner restores v1 → restored bytes baked into v3 dir.
Until LLM approves, live + version_no stay at v2."""
owner_id, owner_cookies = _create_user(web_client, "restoreowner-defer@x.com")
eid = _upload_clean(web_client, owner_cookies, name="restoredefer")
v2 = _make_skill_zip("restoredefer", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
# v2 promoted (guardrails off). Now flip on for the restore.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
r = web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
# version_no must STAY at 2 until LLM approves the v3 (restored)
# copy. version_history has 3 entries; current is still 2.
assert ent["version_no"] == 2, (
f"restore must defer promotion when guardrails on; "
f"version_no={ent['version_no']}"
)
history_n = sorted([int(e["n"]) for e in ent["version_history"]])
assert history_n == [1, 2, 3]
class TestEditPageBanner:
"""Detail page banner during pending edit review must surface
the version under review + the prior version still serving."""
def test_banner_shows_review_error_when_prior_version_still_serving(
self, web_client, monkeypatch,
):
"""v2+ edit landing in review_error must surface a banner to
owner/admin even though entity stays at visibility=approved.
The original gate (visibility != approved) silently hid the
failure — see Bug #2 in plan
when-i-submitted-new-delightful-russell.md."""
from src.repositories.store_submissions import StoreSubmissionsRepository
# Mock LLM to ERROR on v2.
def mock_review_bundle(*args, **kwargs):
return {
"risk_level": None,
"summary": None,
"findings": [],
"template_placeholders_found": 0,
"reviewed_by_model": "mock-model",
"error": "LLMFormatError: mock truncation",
}
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
mock_review_bundle,
)
owner_id, owner_cookies = _create_user(web_client, "errbanner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="errbanner")
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
v2 = _make_skill_zip("errbanner", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Drive the BG review synchronously so the submission row
# lands at review_error.
from src.store_guardrails.runner import run_llm_review
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert ent["visibility_status"] == "approved", (
"deferred promotion: entity stays approved at prior version"
)
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
# Confirm the submission really landed at review_error.
conn = get_system_db()
sub = StoreSubmissionsRepository(conn).get(sub_id)
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert sub["status"] == "review_error"
assert ent["visibility_status"] == "approved", (
"entity must remain approved — prior version still serving"
)
# Detail page MUST surface the failure.
r = web_client.get(
f"/marketplace/flea/{eid}", cookies=owner_cookies,
)
assert r.status_code == 200
body = r.text
# The widened v2+ review_error copy mentions the prior version
# still serving — that's the user-visible signal we just added.
assert "Latest edit failed review" in body, (
"banner partial must render review_error H3 for v2+ edit "
"when prior version still serves"
)
assert "previously approved version (v1)" in body, (
"banner copy must explain why the entity still appears live"
)
# The model's error string must reach the page so the owner
# can see what went wrong.
assert "LLMFormatError" in body
def test_banner_shows_version_n_under_review(self, web_client, monkeypatch):
from src.repositories.store_submissions import StoreSubmissionsRepository
owner_id, owner_cookies = _create_user(web_client, "bannerowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="bannerver")
# Switch guardrails on; stub LLM scheduler so v2 stays pending.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
v2 = _make_skill_zip("bannerver", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
# v2 review pending. visibility on entity stays approved (per
# the deferred-promotion fix). But banner partial reads sub
# status — for an in-flight edit submission the banner won't
# render unless visibility != approved. Lock the scenario:
# ensure entity stays at version_no=1 + visibility approved.
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
sub = StoreSubmissionsRepository(conn).latest_for_entity(eid)
conn.close()
assert ent["version_no"] == 1
assert ent["visibility_status"] == "approved"
assert sub["status"] == "pending_llm"
# Detail page renders. Banner partial only fires when
# visibility_status != approved, so for a deferred-edit case
# the marketplace detail does NOT render the quarantine
# banner — that's correct UX (consumers see the entity as
# approved and operational). Owner-facing review status
# surfaces via the Edit button being disabled.
r = web_client.get(
f"/marketplace/flea/{eid}", cookies=owner_cookies,
)
assert r.status_code == 200
# Edit button should reflect the in-flight review (locked).
assert "review in flight" in r.text or "Edit (review in flight)" in r.text
class TestAuditLogPerVersion:
"""Each edit / restore writes audit rows carrying the version_no
in params, so the entity timeline can attribute events to the
right version."""
def test_edit_audit_carries_version_no(self, web_client):
from src.repositories.audit import AuditRepository
owner_id, owner_cookies = _create_user(web_client, "auditver@x.com")
eid = _upload_clean(web_client, owner_cookies, name="auditver")
v2 = _make_skill_zip("auditver", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
conn = get_system_db()
rows = AuditRepository(conn).query_for_resources(
[f"store_entity:{eid}"], limit=20,
)
conn.close()
# store.entity.update event for the edit must carry version_no
# in its params.
update_rows = [r for r in rows if r.get("action") == "store.entity.update"]
assert update_rows, "missing store.entity.update audit"
assert any(
(r.get("params") or {}).get("version_no") == 2
for r in update_rows
), "update audit must carry version_no=2 in params"
def test_restore_audit_carries_versions(self, web_client):
from src.repositories.audit import AuditRepository
owner_id, owner_cookies = _create_user(web_client, "auditrestore@x.com")
eid = _upload_clean(web_client, owner_cookies, name="auditrest")
v2 = _make_skill_zip("auditrest", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
web_client.post(
f"/api/store/entities/{eid}/versions/1/restore",
cookies=owner_cookies,
)
conn = get_system_db()
rows = AuditRepository(conn).query_for_resources(
[f"store_entity:{eid}"], limit=20,
)
conn.close()
restore_rows = [r for r in rows if r.get("action") == "store.entity.restore"]
assert restore_rows, "missing store.entity.restore audit"
params = restore_rows[0].get("params") or {}
assert params.get("restored_from_version_no") == 1
assert params.get("new_version_no") == 3
class TestPRReviewFixes:
"""Locks in the fixes called out in the PR #239 review."""
def test_block_while_pending_fires_for_v2_edit_under_deferred_promotion(
self, web_client, monkeypatch,
):
"""#1 — v2+ edit during in-flight LLM review must 409 even
though entity.visibility_status is still 'approved'."""
from src.repositories.store_submissions import StoreSubmissionsRepository
owner_id, owner_cookies = _create_user(web_client, "blockv2@x.com")
eid = _upload_clean(web_client, owner_cookies, name="blockv2")
# Switch guardrails on for the edit so promotion defers.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
v2 = _make_skill_zip("blockv2", body="v2 " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Entity should remain at 'approved' visibility — v2 in flight.
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
sub = StoreSubmissionsRepository(conn).latest_for_entity(eid)
conn.close()
assert ent["visibility_status"] == "approved"
assert sub["status"] == "pending_llm"
# Second concurrent edit MUST be blocked.
v3 = _make_skill_zip("blockv2", body="v3 " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v3.zip", v3, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 409, r.text
assert r.json()["detail"]["code"] == "prior_version_pending"
def test_name_change_with_bundle_does_not_rename_live_until_promote(
self, web_client, monkeypatch,
):
"""#2 — name + bundle in same PUT must NOT rename live until
the LLM approves and promotion runs. Existing installer keeps
getting the prior bundle under the prior slug."""
from app.utils import get_store_dir
owner_id, owner_cookies = _create_user(web_client, "rename-defer@x.com")
eid = _upload_clean(web_client, owner_cookies, name="origname")
plugin_dir = Path(get_store_dir()) / eid / "plugin"
old_skill_dir = plugin_dir / "skills" / "origname-by-rename-defer"
assert old_skill_dir.is_dir()
# Enable guardrails so promotion defers.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
v2 = _make_skill_zip("origname", body="v2 " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
data={"name": "newname"},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
# Live skill dir MUST still hold the old slug — promotion
# hasn't fired since we stubbed _schedule_llm_review.
assert old_skill_dir.is_dir(), (
"live skill dir was renamed before LLM approval — "
"violates deferred-promotion contract"
)
new_skill_dir = plugin_dir / "skills" / "newname-by-rename-defer"
assert not new_skill_dir.exists(), (
"live shouldn't have the renamed slug yet"
)
# Version dir HAS been renamed (so promotion will land on the
# new slug).
v2_dir = (
Path(get_store_dir()) / eid / "versions" / "v2" / "plugin"
/ "skills" / "newname-by-rename-defer"
)
assert v2_dir.is_dir(), "version dir should carry the new slug"
def test_v2_approval_logs_approved_not_skipped(
self, web_client, monkeypatch,
):
"""#3 — v2+ approvals must log store.submission.approved, NOT
store.submission.bg_verdict_skipped. Pre-fix the runner used
the visibility-flip return value to gate the audit; under
deferred promotion v2+ never flips visibility (already
'approved'), so the wrong audit was emitted."""
from src.repositories.audit import AuditRepository
from src.repositories.store_submissions import StoreSubmissionsRepository
from src.store_guardrails.runner import run_llm_review
from app.utils import get_store_dir
owner_id, owner_cookies = _create_user(web_client, "auditv2@x.com")
eid = _upload_clean(web_client, owner_cookies, name="auditv2")
# Enable guardrails for the edit. Stub LLM scheduler so we
# control when the runner fires.
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
# Mock the LLM call to return a safe verdict.
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
lambda *a, **kw: {
"risk_level": "safe", "summary": "ok",
"findings": [], "template_placeholders_found": 0,
"reviewed_by_model": "mock", "error": None,
},
)
v2 = _make_skill_zip("auditv2", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
# Manually fire the runner against the v2 dir.
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
# Audit log must contain store.submission.approved with
# promoted_to_version_no=2; NO bg_verdict_skipped.
conn = get_system_db()
rows = AuditRepository(conn).query_for_resources(
[f"store_submission:{sub_id}"], limit=20,
)
conn.close()
actions = [r.get("action") for r in rows]
assert "store.submission.approved" in actions, (
f"v2 approval missing approved audit; got {actions}"
)
assert "store.submission.bg_verdict_skipped" not in actions, (
f"v2 approval should NOT log bg_verdict_skipped; got {actions}"
)
approved_row = next(
r for r in rows if r.get("action") == "store.submission.approved"
)
params = approved_row.get("params") or {}
assert params.get("promoted_to_version_no") == 2
def test_bg_verdict_skipped_when_admin_archives_during_review(
self, web_client, monkeypatch,
):
"""Negative: when admin DOES archive mid-review, the runner
correctly logs bg_verdict_skipped (not approved)."""
from src.repositories.audit import AuditRepository
from src.repositories.store_submissions import StoreSubmissionsRepository
from src.repositories.store_entities import StoreEntitiesRepository
from src.store_guardrails.runner import run_llm_review
from app.utils import get_store_dir
owner_id, owner_cookies = _create_user(web_client, "archmid@x.com")
eid = _upload_clean(web_client, owner_cookies, name="archmid")
monkeypatch.setattr("app.api.store.get_guardrails_enabled", lambda: True)
monkeypatch.setattr("app.api.store.get_guardrails_llm_provider_ready", lambda: True)
monkeypatch.setattr(
"app.api.store._schedule_llm_review",
lambda *a, **kw: None,
)
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
lambda *a, **kw: {
"risk_level": "safe", "summary": "ok",
"findings": [], "template_placeholders_found": 0,
"reviewed_by_model": "mock", "error": None,
},
)
v2 = _make_skill_zip("archmid", body="v2 " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
# Admin archives BEFORE runner fires.
conn = get_system_db()
StoreEntitiesRepository(conn).archive(eid, by_user_id="admin-x")
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
run_llm_review(
sub_id,
plugin_dir=Path(get_store_dir()) / eid / "versions" / "v2" / "plugin",
conn_factory=get_system_db,
api_key_loader=lambda: "sk-test",
model_loader=lambda: "claude-haiku-4-5-20251001",
)
conn = get_system_db()
rows = AuditRepository(conn).query_for_resources(
[f"store_submission:{sub_id}"], limit=20,
)
conn.close()
actions = [r.get("action") for r in rows]
assert "store.submission.bg_verdict_skipped" in actions, (
f"archive-during-review must log bg_verdict_skipped; got {actions}"
)
assert "store.submission.approved" not in actions, (
f"archive-during-review must NOT log approved; got {actions}"
)
class TestAdminQueueShowsVersion:
def test_admin_queue_shows_v_no_after_name(self, web_client):
"""v# column derives version_no from entity.version_history by
matching submission.version (hash) against the entry hashes."""
owner_id, owner_cookies = _create_user(web_client, "vqowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="vqcol")
v2 = _make_skill_zip("vqcol", body="v2 body. " * 80)
web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
_, admin_cookies = _create_admin(web_client)
r = web_client.get(
"/api/admin/store/submissions", cookies=admin_cookies,
)
assert r.status_code == 200
items = {it["name"]: it for it in r.json()["items"]}
assert "vqcol" in items
# version_no derived for the v2 row should be 2.
# The list returns rows newest-first; pick the v2 (current).
v2_row = next(
it for it in r.json()["items"]
if it.get("entity_id") == eid and it.get("version_no") == 2
)
assert v2_row["version_no"] == 2
assert v2_row["entity_version_no"] == 2
def test_admin_detail_shows_version_no(self, web_client):
"""Detail page renders v# under the Status / Entity-lifecycle
block + a separate Bundle hash row."""
from src.repositories.store_submissions import StoreSubmissionsRepository
owner_id, owner_cookies = _create_user(web_client, "vdowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="vdrow")
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
_, admin_cookies = _create_admin(web_client)
r = web_client.get(
f"/admin/store/submissions/{sub_id}", cookies=admin_cookies,
)
assert r.status_code == 200
body = r.text
assert "<dt>Version</dt>" in body
assert "<dt>Bundle hash</dt>" in body
assert "v1" in body
class TestPublishGateFailClosed:
"""Hold-for-review when ``guardrails.enabled: true`` but no LLM
provider credentials are present in env. The pre-v45 fall-back
silently auto-approved every upload — a fail-OPEN hole the
operator couldn't notice. New behavior: submissions sit at
``pending_llm``, entity stays at ``visibility_status='pending'``,
admin retries from /admin/store/submissions after providing
credentials."""
def test_v1_upload_enabled_but_not_ready_holds_at_pending(
self, web_client, monkeypatch,
):
from src.repositories.store_submissions import StoreSubmissionsRepository
# Flip guardrails ON but leave provider_ready as False.
monkeypatch.setattr(
"app.api.store.get_guardrails_enabled", lambda: True,
)
monkeypatch.setattr(
"app.api.store.get_guardrails_llm_provider_ready", lambda: False,
)
# No mock review_bundle — we should never call the LLM.
# If we did, the lack of patching would surface as a real
# network call attempt, easy to catch as a hang.
owner_id, owner_cookies = _create_user(web_client, "holdv1@x.com")
eid = _upload_clean(web_client, owner_cookies, name="holdv1")
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
sub = StoreSubmissionsRepository(conn).latest_for_entity(eid)
conn.close()
assert ent["visibility_status"] == "pending", (
"enabled-but-not-ready must NOT publish — entity stays pending"
)
assert sub["status"] == "pending_llm", (
"submission must hold at pending_llm awaiting admin retry"
)
assert sub["llm_findings"] is None, (
"no LLM call was made — findings must be empty"
)
def test_admin_retry_pending_llm_fires_review(
self, web_client, monkeypatch,
):
"""After the operator sets the API key, admin Retry-review on a
held pending_llm row schedules + runs the LLM."""
from src.repositories.store_submissions import StoreSubmissionsRepository
# Phase 1: upload with provider not-ready → held at pending_llm.
monkeypatch.setattr(
"app.api.store.get_guardrails_enabled", lambda: True,
)
monkeypatch.setattr(
"app.api.store.get_guardrails_llm_provider_ready", lambda: False,
)
owner_id, owner_cookies = _create_user(web_client, "retryholder@x.com")
eid = _upload_clean(web_client, owner_cookies, name="retryholder")
conn = get_system_db()
sub_id = StoreSubmissionsRepository(conn).latest_for_entity(eid)["id"]
conn.close()
# Phase 2: operator adds credentials, admin retries.
# Inject a fake env var so default_api_key_loader doesn't raise
# before the mock review_bundle runs.
monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-test-fake-key-for-retry")
# Mock review_bundle so the retry resolves to approved without
# touching the network.
def mock_review_bundle(*args, **kwargs):
return {
"risk_level": "safe", "summary": "ok",
"findings": [], "template_placeholders_found": 0,
"reviewed_by_model": "mock-model", "error": None,
"content_quality": {"verdict": "pass", "issues": []},
}
monkeypatch.setattr(
"src.store_guardrails.llm_review.review_bundle",
mock_review_bundle,
)
_, admin_cookies = _create_admin(web_client)
r = web_client.post(
f"/api/admin/store/submissions/{sub_id}/retry",
cookies=admin_cookies,
)
assert r.status_code == 200, r.text
# After retry, BG task runs synchronously in TestClient (it
# blocks the response). Verify the row moved to approved.
conn = get_system_db()
sub = StoreSubmissionsRepository(conn).get(sub_id)
ent = StoreEntitiesRepository(conn).get(eid)
conn.close()
assert sub["status"] == "approved", (
f"retry must drive submission to approved; got {sub['status']}"
)
assert ent["visibility_status"] == "approved", (
"entity must flip to approved after LLM ok"
)
def test_edit_enabled_but_not_ready_holds_prior_serving(
self, web_client, monkeypatch,
):
"""v2+ edit under enabled-but-not-ready: v1 keeps serving,
v2 submission held at pending_llm. Critical safety property:
no silent promotion."""
from src.repositories.store_submissions import StoreSubmissionsRepository
# Initial upload runs with guardrails OFF (autouse default) →
# v1 approved. Then flip to enabled-but-not-ready for PUT.
owner_id, owner_cookies = _create_user(web_client, "holdedit@x.com")
eid = _upload_clean(web_client, owner_cookies, name="holdedit")
conn = get_system_db()
v1_hash = StoreEntitiesRepository(conn).get(eid)["version"]
conn.close()
monkeypatch.setattr(
"app.api.store.get_guardrails_enabled", lambda: True,
)
monkeypatch.setattr(
"app.api.store.get_guardrails_llm_provider_ready", lambda: False,
)
v2 = _make_skill_zip("holdedit", body="V2 BODY " * 80)
r = web_client.put(
f"/api/store/entities/{eid}",
files={"file": ("v2.zip", v2, "application/zip")},
cookies=owner_cookies,
)
assert r.status_code == 200, r.text
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
sub = StoreSubmissionsRepository(conn).latest_for_entity(eid)
conn.close()
# Entity stays approved at v1, v2 sits at pending_llm.
assert ent["visibility_status"] == "approved"
assert ent["version_no"] == 1
assert ent["version"] == v1_hash, (
"live bundle must remain v1 — no silent promotion of v2"
)
assert sub["status"] == "pending_llm"
assert sub["llm_findings"] is None
def test_disabled_intent_still_auto_approves(
self, web_client, monkeypatch,
):
"""Operator explicitly opting out (``enabled: false``) keeps
the prior auto-approve behavior — local dev / no-LLM
deployments aren't blocked."""
from src.repositories.store_submissions import StoreSubmissionsRepository
# autouse fixture already sets enabled=False. Just confirm
# behavior end-to-end.
owner_id, owner_cookies = _create_user(web_client, "offowner@x.com")
eid = _upload_clean(web_client, owner_cookies, name="offowner")
conn = get_system_db()
ent = StoreEntitiesRepository(conn).get(eid)
sub = StoreSubmissionsRepository(conn).latest_for_entity(eid)
conn.close()
assert ent["visibility_status"] == "approved"
assert sub["status"] == "approved"