@openwop/openwop-conformance 1.5.0 → 1.6.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +27 -0
- package/README.md +2 -2
- package/api/asyncapi.yaml +25 -4
- package/api/openapi.yaml +371 -0
- package/coverage.md +31 -4
- package/fixtures/conformance-phase4-nondet-tool.json +53 -0
- package/fixtures/conformance-phase4-replay-divergence.json +40 -0
- package/fixtures.md +5 -3
- package/package.json +1 -1
- package/schemas/README.md +4 -0
- package/schemas/annotation-create.schema.json +37 -0
- package/schemas/annotation.schema.json +56 -0
- package/schemas/capabilities.schema.json +191 -3
- package/schemas/credential-reference.schema.json +21 -0
- package/schemas/node-pack-manifest.schema.json +112 -1
- package/schemas/run-diff-response.schema.json +64 -0
- package/schemas/run-event-payloads.schema.json +104 -2
- package/schemas/run-event.schema.json +8 -1
- package/schemas/run-snapshot.schema.json +11 -0
- package/src/lib/behavior-gate.ts +51 -0
- package/src/lib/driver.ts +13 -1
- package/src/lib/feedback.ts +31 -0
- package/src/lib/saml-idp.ts +179 -0
- package/src/scenarios/approval-gate-events.test.ts +61 -0
- package/src/scenarios/approval-gate-flow.test.ts +68 -0
- package/src/scenarios/auth-saml-profile.test.ts +119 -0
- package/src/scenarios/auth-scim-profile.test.ts +65 -0
- package/src/scenarios/authorization-fail-closed.test.ts +80 -0
- package/src/scenarios/authorization-roles-shape.test.ts +83 -0
- package/src/scenarios/connector-manifest-validity.test.ts +142 -0
- package/src/scenarios/credential-payload-redaction.test.ts +93 -0
- package/src/scenarios/credentials-capability-shape.test.ts +90 -0
- package/src/scenarios/cross-engine-append-behavior.test.ts +204 -0
- package/src/scenarios/cross-host-traceparent-propagation.test.ts +13 -6
- package/src/scenarios/cross-workspace-isolation.test.ts +72 -0
- package/src/scenarios/deadletter-capability-shape.test.ts +59 -0
- package/src/scenarios/deadletter-retry-exhaustion.test.ts +62 -0
- package/src/scenarios/experimental-tier-shape.test.ts +192 -0
- package/src/scenarios/feedback-capability-shape.test.ts +35 -0
- package/src/scenarios/feedback-correction-redaction.test.ts +35 -0
- package/src/scenarios/feedback-cross-tenant-isolation.test.ts +37 -0
- package/src/scenarios/feedback-fork-not-copied.test.ts +40 -0
- package/src/scenarios/feedback-on-terminal-run.test.ts +32 -0
- package/src/scenarios/feedback-record-and-list.test.ts +32 -0
- package/src/scenarios/feedback-unsupported-501.test.ts +32 -0
- package/src/scenarios/identity-owner-shape.test.ts +64 -0
- package/src/scenarios/multi-agent-confidence-escalation.test.ts +13 -12
- package/src/scenarios/multi-agent-memory-lifecycle.test.ts +87 -12
- package/src/scenarios/multi-region-idempotency-behavior.test.ts +203 -0
- package/src/scenarios/oauth-capability-shape.test.ts +97 -0
- package/src/scenarios/oauth-connector-redaction.test.ts +91 -0
- package/src/scenarios/pack-registry-isolation.test.ts +108 -0
- package/src/scenarios/pack-registry-publish.test.ts +1 -1
- package/src/scenarios/prompt-mutation-workspace-membership-enforced.test.ts +126 -0
- package/src/scenarios/prompt-read-workspace-membership-enforced.test.ts +183 -0
- package/src/scenarios/redaction.test.ts +4 -1
- package/src/scenarios/replay-divergence-at-refusal.test.ts +187 -7
- package/src/scenarios/replay-observable-sequence-determinism.test.ts +20 -6
- package/src/scenarios/run-diff.test.ts +143 -0
- package/src/scenarios/sandbox-capability-gate-respected.test.ts +7 -1
- package/src/scenarios/sandbox-memory-cap.test.ts +7 -5
- package/src/scenarios/sandbox-mvp-behavior.test.ts +280 -0
- package/src/scenarios/sandbox-no-cross-pack-mutation.test.ts +7 -1
- package/src/scenarios/sandbox-no-host-env-leak.test.ts +5 -1
- package/src/scenarios/sandbox-no-host-fs-escape.test.ts +9 -1
- package/src/scenarios/sandbox-no-host-process-escape.test.ts +5 -1
- package/src/scenarios/sandbox-no-network-escape.test.ts +5 -1
- package/src/scenarios/sandbox-timeout-cap.test.ts +7 -5
- package/src/scenarios/scheduling-capability-shape.test.ts +81 -0
- package/src/scenarios/scheduling-cron-fires-once.test.ts +66 -0
- package/src/scenarios/secret-leakage-otel-attribute.test.ts +241 -0
- package/src/scenarios/spec-corpus-validity.test.ts +6 -3
|
@@ -0,0 +1,97 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* oauth-capability-shape — RFC 0047 §A advertisement-shape verification.
|
|
3
|
+
*
|
|
4
|
+
* Status: DRAFT. RFC 0047 (`host.oauth`) is `Draft`. The
|
|
5
|
+
* `capabilities.oauth` block has landed in `schemas/capabilities.schema.json`.
|
|
6
|
+
*
|
|
7
|
+
* Always runs (shape-only): when the host advertises `capabilities.oauth`,
|
|
8
|
+
* its fields MUST be well-formed; when it doesn't, the block is absent.
|
|
9
|
+
*
|
|
10
|
+
* What this scenario asserts:
|
|
11
|
+
* 1. `capabilities.oauth` is either absent or a well-formed object.
|
|
12
|
+
* 2. When `supported: true`, `grants` (when present) is a subset of
|
|
13
|
+
* {authorization_code, client_credentials, refresh_token}, and every
|
|
14
|
+
* `providers[]` entry has a non-empty `id` (RFC 0047 §A).
|
|
15
|
+
*
|
|
16
|
+
* @see RFCS/0047-host-oauth-connector-flows.md
|
|
17
|
+
*/
|
|
18
|
+
|
|
19
|
+
import { describe, it, expect } from 'vitest';
|
|
20
|
+
import { driver } from '../lib/driver.js';
|
|
21
|
+
|
|
22
|
+
interface DiscoveryOAuthProvider {
|
|
23
|
+
id?: string;
|
|
24
|
+
authUrl?: string;
|
|
25
|
+
tokenUrl?: string;
|
|
26
|
+
scopesSupported?: string[];
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
interface DiscoveryOAuth {
|
|
30
|
+
supported?: boolean;
|
|
31
|
+
grants?: string[];
|
|
32
|
+
providers?: DiscoveryOAuthProvider[];
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
interface DiscoveryDoc {
|
|
36
|
+
capabilities?: {
|
|
37
|
+
oauth?: DiscoveryOAuth;
|
|
38
|
+
};
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
const VALID_GRANTS: ReadonlySet<string> = new Set([
|
|
42
|
+
'authorization_code',
|
|
43
|
+
'client_credentials',
|
|
44
|
+
'refresh_token',
|
|
45
|
+
]);
|
|
46
|
+
|
|
47
|
+
async function readOAuth(): Promise<DiscoveryOAuth | null> {
|
|
48
|
+
const res = await driver.get('/.well-known/openwop');
|
|
49
|
+
const body = res.json as DiscoveryDoc | undefined;
|
|
50
|
+
return body?.capabilities?.oauth ?? null;
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
describe('oauth-capability-shape: advertisement shape (RFC 0047 §A)', () => {
|
|
54
|
+
it('capabilities.oauth is either absent or well-formed', async () => {
|
|
55
|
+
const oauth = await readOAuth();
|
|
56
|
+
if (oauth === null) return; // host doesn't advertise host.oauth at all
|
|
57
|
+
expect(
|
|
58
|
+
typeof oauth.supported,
|
|
59
|
+
driver.describe(
|
|
60
|
+
'capabilities.schema.json §oauth',
|
|
61
|
+
'capabilities.oauth.supported MUST be a boolean when oauth is advertised',
|
|
62
|
+
),
|
|
63
|
+
).toBe('boolean');
|
|
64
|
+
});
|
|
65
|
+
|
|
66
|
+
it('grants is a subset of the canonical grant set when supported', async () => {
|
|
67
|
+
const oauth = await readOAuth();
|
|
68
|
+
if (!oauth?.supported || oauth.grants === undefined) return;
|
|
69
|
+
expect(
|
|
70
|
+
Array.isArray(oauth.grants),
|
|
71
|
+
driver.describe('RFC 0047 §A', 'capabilities.oauth.grants MUST be an array'),
|
|
72
|
+
).toBe(true);
|
|
73
|
+
for (const grant of oauth.grants) {
|
|
74
|
+
expect(
|
|
75
|
+
VALID_GRANTS.has(grant),
|
|
76
|
+
driver.describe(
|
|
77
|
+
'RFC 0047 §A',
|
|
78
|
+
`capabilities.oauth.grants entries MUST be one of {${[...VALID_GRANTS].join(', ')}}, got: ${grant}`,
|
|
79
|
+
),
|
|
80
|
+
).toBe(true);
|
|
81
|
+
}
|
|
82
|
+
});
|
|
83
|
+
|
|
84
|
+
it('every advertised provider has a non-empty id when supported', async () => {
|
|
85
|
+
const oauth = await readOAuth();
|
|
86
|
+
if (!oauth?.supported || oauth.providers === undefined) return;
|
|
87
|
+
for (const provider of oauth.providers) {
|
|
88
|
+
expect(
|
|
89
|
+
typeof provider.id === 'string' && provider.id.length > 0,
|
|
90
|
+
driver.describe(
|
|
91
|
+
'RFC 0047 §A',
|
|
92
|
+
'each capabilities.oauth.providers[] entry MUST declare a non-empty id',
|
|
93
|
+
),
|
|
94
|
+
).toBe(true);
|
|
95
|
+
}
|
|
96
|
+
});
|
|
97
|
+
});
|
|
@@ -0,0 +1,91 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* oauth-connector-redaction — RFC 0047 §C / §D + `credential-payload-redaction`.
|
|
3
|
+
*
|
|
4
|
+
* Status: DRAFT. RFC 0047 (`host.oauth`) is `Draft`. Reuses the RFC 0046
|
|
5
|
+
* SECURITY invariant `credential-payload-redaction` — OAuth tokens acquired
|
|
6
|
+
* via host.oauth are stored as host.credentials entries and are subject to
|
|
7
|
+
* the same no-plaintext-on-the-wire rule.
|
|
8
|
+
*
|
|
9
|
+
* Capability-gated: skips when the host does not advertise
|
|
10
|
+
* `capabilities.oauth.supported = true`.
|
|
11
|
+
*
|
|
12
|
+
* What this scenario asserts:
|
|
13
|
+
* 1. Advertisement shape — `capabilities.oauth.supported` is a boolean.
|
|
14
|
+
* 2. Token-material redaction MUST-NOT — when the host exposes the optional
|
|
15
|
+
* `POST /v1/host/sample/oauth/connector-echo` test seam (a synthetic
|
|
16
|
+
* provider acquires a token whose value is a known canary, then a
|
|
17
|
+
* connector node runs), the canary MUST NOT appear in ANY observable
|
|
18
|
+
* run surface, and `connector.authorized` MUST carry the credential
|
|
19
|
+
* reference rather than the token.
|
|
20
|
+
*
|
|
21
|
+
* Hosts without the seam soft-skip the redaction probe (404).
|
|
22
|
+
*
|
|
23
|
+
* @see RFCS/0047-host-oauth-connector-flows.md
|
|
24
|
+
* @see SECURITY/invariants.yaml id: credential-payload-redaction
|
|
25
|
+
*/
|
|
26
|
+
|
|
27
|
+
import { describe, it, expect } from 'vitest';
|
|
28
|
+
import { driver } from '../lib/driver.js';
|
|
29
|
+
|
|
30
|
+
interface DiscoveryOAuth {
|
|
31
|
+
supported?: boolean;
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
interface DiscoveryDoc {
|
|
35
|
+
capabilities?: {
|
|
36
|
+
oauth?: DiscoveryOAuth;
|
|
37
|
+
};
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
const TOKEN_CANARY = 'OPENWOP_OAUTH_CANARY_b7d3e1a9c2';
|
|
41
|
+
|
|
42
|
+
async function readOAuth(): Promise<DiscoveryOAuth | null> {
|
|
43
|
+
const res = await driver.get('/.well-known/openwop');
|
|
44
|
+
const body = res.json as DiscoveryDoc | undefined;
|
|
45
|
+
return body?.capabilities?.oauth ?? null;
|
|
46
|
+
}
|
|
47
|
+
|
|
48
|
+
describe('oauth-connector-redaction: advertisement shape (RFC 0047 §A)', () => {
|
|
49
|
+
it('capabilities.oauth.supported is a boolean when advertised', async () => {
|
|
50
|
+
const oauth = await readOAuth();
|
|
51
|
+
if (oauth === null) return;
|
|
52
|
+
expect(
|
|
53
|
+
typeof oauth.supported,
|
|
54
|
+
driver.describe(
|
|
55
|
+
'capabilities.schema.json §oauth',
|
|
56
|
+
'capabilities.oauth.supported MUST be a boolean when oauth is advertised',
|
|
57
|
+
),
|
|
58
|
+
).toBe('boolean');
|
|
59
|
+
});
|
|
60
|
+
});
|
|
61
|
+
|
|
62
|
+
describe('oauth-connector-redaction: token material MUST NOT cross the wire (RFC 0047 §C.2)', () => {
|
|
63
|
+
it('canary token is absent from every observable run surface', async () => {
|
|
64
|
+
const oauth = await readOAuth();
|
|
65
|
+
if (!oauth?.supported) return; // capability-gated
|
|
66
|
+
|
|
67
|
+
// Seam contract: a synthetic provider issues a token whose value is
|
|
68
|
+
// TOKEN_CANARY, a connector node runs, and the run's observable surfaces
|
|
69
|
+
// (events incl. connector.authorized + snapshot + debug bundle) are returned.
|
|
70
|
+
const res = await driver.post('/v1/host/sample/oauth/connector-echo', { canary: TOKEN_CANARY });
|
|
71
|
+
// 404 from a host that hasn't wired the test seam is a soft-skip.
|
|
72
|
+
if (res.status === 404) return;
|
|
73
|
+
|
|
74
|
+
expect(
|
|
75
|
+
res.status,
|
|
76
|
+
driver.describe(
|
|
77
|
+
'RFC 0047 §C',
|
|
78
|
+
'the oauth connector-echo seam MUST acquire the token and return the run observable surfaces',
|
|
79
|
+
),
|
|
80
|
+
).toBeLessThan(400);
|
|
81
|
+
|
|
82
|
+
const serialized = JSON.stringify(res.json ?? {});
|
|
83
|
+
expect(
|
|
84
|
+
serialized.includes(TOKEN_CANARY),
|
|
85
|
+
driver.describe(
|
|
86
|
+
'SECURITY/invariants.yaml credential-payload-redaction',
|
|
87
|
+
'acquired OAuth token material MUST NOT appear in inputs, variables, channels, events, snapshot, or debug bundle — only the credential reference may cross the wire',
|
|
88
|
+
),
|
|
89
|
+
).toBe(false);
|
|
90
|
+
});
|
|
91
|
+
});
|
|
@@ -0,0 +1,108 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Pack-registry test-mode isolation — RFC 0025 §C point 1.
|
|
3
|
+
*
|
|
4
|
+
* Status: BEHAVIORAL (soft-skip). A pack PUT'd to `/v1/packs-test/*` MUST
|
|
5
|
+
* NOT appear in `/v1/packs/*` listings. This anchors the test-mode
|
|
6
|
+
* mirror's load-bearing safety invariant: the conformance suite is
|
|
7
|
+
* trusted to drive publish-error-catalog traffic against the test
|
|
8
|
+
* namespace precisely because the test catalog is guaranteed distinct
|
|
9
|
+
* from the production catalog.
|
|
10
|
+
*
|
|
11
|
+
* Soft-skips when the host doesn't advertise
|
|
12
|
+
* `capabilities.packs.testMode.supported: true` (or advertises
|
|
13
|
+
* `isolated: false` — in which case the host is honestly disclaiming
|
|
14
|
+
* the invariant and the conformance suite's other publish-error tests
|
|
15
|
+
* are not applicable either).
|
|
16
|
+
*
|
|
17
|
+
* @see RFCS/0025-test-mode-registry-namespace.md §C "Isolation guarantees"
|
|
18
|
+
* @see schemas/capabilities.schema.json §packs.testMode
|
|
19
|
+
* @see pack-registry-publish.test.ts (the 25 sibling scenarios this invariant unblocks)
|
|
20
|
+
*/
|
|
21
|
+
|
|
22
|
+
import { describe, it, expect } from 'vitest';
|
|
23
|
+
import { driver } from '../lib/driver.js';
|
|
24
|
+
|
|
25
|
+
interface DiscoveryDoc {
|
|
26
|
+
capabilities?: Record<string, unknown>;
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
interface TestModeAdvertisement {
|
|
30
|
+
readonly supported: boolean;
|
|
31
|
+
readonly isolated: boolean;
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
async function getTestModeAdvertisement(): Promise<TestModeAdvertisement | null> {
|
|
35
|
+
const res = await driver.get('/.well-known/openwop');
|
|
36
|
+
const body = res.json as DiscoveryDoc | undefined;
|
|
37
|
+
const top = body?.capabilities as Record<string, unknown> | undefined;
|
|
38
|
+
const packs = top && typeof top === 'object' ? (top['packs'] as Record<string, unknown> | undefined) : undefined;
|
|
39
|
+
const testMode = packs && typeof packs === 'object' ? (packs['testMode'] as Record<string, unknown> | undefined) : undefined;
|
|
40
|
+
if (!testMode || typeof testMode !== 'object') return null;
|
|
41
|
+
return {
|
|
42
|
+
supported: testMode['supported'] === true,
|
|
43
|
+
isolated: testMode['isolated'] === true,
|
|
44
|
+
};
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
function freshPackName(): string {
|
|
48
|
+
return `core.openwop.test-isolation-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
describe('pack-registry-isolation: test catalog MUST NOT bleed into production (RFC 0025 §C.1)', () => {
|
|
52
|
+
it('a pack PUT to /v1/packs-test/{name} MUST NOT appear in GET /v1/packs/{name}', async () => {
|
|
53
|
+
const adv = await getTestModeAdvertisement();
|
|
54
|
+
if (!adv || !adv.supported) return; // host doesn't advertise the seam
|
|
55
|
+
if (!adv.isolated) return; // host explicitly disclaims the invariant — no contract to assert
|
|
56
|
+
|
|
57
|
+
const name = freshPackName();
|
|
58
|
+
const version = '1.0.0';
|
|
59
|
+
|
|
60
|
+
// PUT to the test namespace. The body is intentionally minimal — the
|
|
61
|
+
// isolation invariant is independent of whether validation accepts
|
|
62
|
+
// or rejects the publish. Either outcome is fine; what's tested is
|
|
63
|
+
// that NEITHER outcome causes the pack to surface in the production
|
|
64
|
+
// catalog.
|
|
65
|
+
const putRes = await driver.put(
|
|
66
|
+
`/v1/packs-test/${encodeURIComponent(name)}/-/${encodeURIComponent(version)}.tgz`,
|
|
67
|
+
Buffer.from([0x1f, 0x8b, 0]),
|
|
68
|
+
{ headers: { 'Content-Type': 'application/octet-stream' } },
|
|
69
|
+
);
|
|
70
|
+
|
|
71
|
+
// If the seam returns 404, the test-mode endpoint isn't actually
|
|
72
|
+
// wired up despite the advertisement — pack-registry-publish.test.ts
|
|
73
|
+
// catches that drift in 24 other scenarios; soft-skip here.
|
|
74
|
+
if (putRes.status === 404) return;
|
|
75
|
+
|
|
76
|
+
// Probe the production namespace. The invariant: a pack written
|
|
77
|
+
// via /v1/packs-test/* MUST NOT be retrievable via /v1/packs/*.
|
|
78
|
+
const prodRes = await driver.get(`/v1/packs/${encodeURIComponent(name)}`);
|
|
79
|
+
|
|
80
|
+
// 404 is the canonical "not found" — exactly what isolation requires.
|
|
81
|
+
// 200 with a payload that does NOT name our pack would mean the host
|
|
82
|
+
// returned a listing of unrelated packs (some hosts serve search-shaped
|
|
83
|
+
// results on /v1/packs/{nonexistent}); we check the negative explicitly.
|
|
84
|
+
if (prodRes.status === 200) {
|
|
85
|
+
const body = prodRes.json as Record<string, unknown> | undefined;
|
|
86
|
+
const stringified = body ? JSON.stringify(body) : '';
|
|
87
|
+
expect(
|
|
88
|
+
stringified.includes(name),
|
|
89
|
+
driver.describe(
|
|
90
|
+
'RFCS/0025-test-mode-registry-namespace.md §C point 1',
|
|
91
|
+
`pack name '${name}' was written via /v1/packs-test/${name}@${version} but appeared in /v1/packs/${name} response body — test-catalog isolation MUST hold`,
|
|
92
|
+
),
|
|
93
|
+
).toBe(false);
|
|
94
|
+
return;
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
// Acceptable: 4xx range (404 pack_not_found is the spec-canonical
|
|
98
|
+
// shape; 410/422 also fine — any "not present in production catalog"
|
|
99
|
+
// signal satisfies the invariant).
|
|
100
|
+
expect(
|
|
101
|
+
prodRes.status >= 400 && prodRes.status < 500,
|
|
102
|
+
driver.describe(
|
|
103
|
+
'RFCS/0025-test-mode-registry-namespace.md §C point 1',
|
|
104
|
+
`expected production-namespace GET to return 4xx for a test-namespace-only pack '${name}', got ${prodRes.status}`,
|
|
105
|
+
),
|
|
106
|
+
).toBe(true);
|
|
107
|
+
});
|
|
108
|
+
});
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
/**
|
|
2
2
|
* Pack-registry publish scenarios — `node-packs.md` §"PUT /v1/packs/{name}/-/{version}.tgz".
|
|
3
3
|
*
|
|
4
|
-
* Status: BEHAVIORAL (soft-skip). Per RFC 0025 (`
|
|
4
|
+
* Status: BEHAVIORAL (soft-skip). Per RFC 0025 (`Active` 2026-05-19),
|
|
5
5
|
* the conformance suite drives the documented 19-code error catalog
|
|
6
6
|
* via the test-mode mirror namespace `/v1/packs-test/*`, gated on
|
|
7
7
|
* `capabilities.packs.testMode.supported: true`. Each scenario soft-
|
|
@@ -0,0 +1,126 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* prompt-mutation-workspace-membership-enforced — RFC 0028 Tier-2 §"Workspace
|
|
3
|
+
* membership on workspace-scoped writes" verification.
|
|
4
|
+
*
|
|
5
|
+
* Status: ACTIVE (capability-gated; behavioral when the host advertises
|
|
6
|
+
* `capabilities.prompts.mutableLibrary: true`). Hosts that don't advertise
|
|
7
|
+
* mutableLibrary soft-skip cleanly.
|
|
8
|
+
*
|
|
9
|
+
* The contract (spec/v1/prompts.md §"Discovery & distribution" §"REST
|
|
10
|
+
* endpoints" §"Workspace membership on workspace-scoped writes"):
|
|
11
|
+
*
|
|
12
|
+
* Hosts MUST verify that the authenticated principal is a member of the
|
|
13
|
+
* target workspace BEFORE honoring any POST / PUT / DELETE to a
|
|
14
|
+
* workspace-scoped /v1/prompts* resource. A workspaceId supplied by the
|
|
15
|
+
* caller (request body, URL, or query string) MUST NOT be trusted as
|
|
16
|
+
* authorization on its own. Non-members MUST be rejected fail-closed
|
|
17
|
+
* (typically 403) before any persistence occurs.
|
|
18
|
+
*
|
|
19
|
+
* The probe drives `POST /v1/prompts` with a `workspaceId` the conformance
|
|
20
|
+
* principal cannot be a member of (a cryptographically-unique random value
|
|
21
|
+
* by default; operator-overridable via `OPENWOP_TEST_NONMEMBER_WORKSPACE_ID`
|
|
22
|
+
* for hosts that need a specific synthetic workspace shape). The behavioral
|
|
23
|
+
* MUST is that the host refuses — NOT a 2xx. Any 4xx/5xx is acceptable
|
|
24
|
+
* (401 = auth not configured for this surface; 403 = membership check;
|
|
25
|
+
* 404 = endpoint absent; 422 = body validation; 501 = capability not
|
|
26
|
+
* provided). The failure mode this invariant guards against is a SILENT
|
|
27
|
+
* 2xx with a write to a workspace the caller doesn't belong to — that's the
|
|
28
|
+
* RFC 0028 Tier-2 vulnerability self-disclosed by an adopter on 2026-05-25.
|
|
29
|
+
*
|
|
30
|
+
* Why a random workspaceId is sufficient: a non-member workspace check is
|
|
31
|
+
* negative-space — the host MUST refuse for ANY workspace the principal
|
|
32
|
+
* isn't a member of, and a random UUID has astronomically-low collision
|
|
33
|
+
* probability with any real workspace membership grant.
|
|
34
|
+
*
|
|
35
|
+
* @see RFCS/0028-prompt-library-endpoints.md §"Post-promotion notes"
|
|
36
|
+
* @see spec/v1/prompts.md §"Security invariants" §prompt-mutation-workspace-membership-enforced
|
|
37
|
+
* @see spec/v1/auth.md §"Identity claims — tenant · workspace · principal"
|
|
38
|
+
* @see RFCS/0048-tenant-workspace-principal-identity-model.md §D
|
|
39
|
+
*/
|
|
40
|
+
|
|
41
|
+
import { describe, it, expect } from 'vitest';
|
|
42
|
+
import { randomUUID } from 'node:crypto';
|
|
43
|
+
import { driver } from '../lib/driver.js';
|
|
44
|
+
|
|
45
|
+
const HTTP_SKIP = !process.env.OPENWOP_BASE_URL;
|
|
46
|
+
|
|
47
|
+
interface DiscoveryDoc {
|
|
48
|
+
capabilities?: {
|
|
49
|
+
prompts?: {
|
|
50
|
+
mutableLibrary?: unknown;
|
|
51
|
+
};
|
|
52
|
+
};
|
|
53
|
+
}
|
|
54
|
+
|
|
55
|
+
async function readDiscovery(): Promise<DiscoveryDoc | null> {
|
|
56
|
+
try {
|
|
57
|
+
const res = await driver.get('/.well-known/openwop');
|
|
58
|
+
if (res.status !== 200) return null;
|
|
59
|
+
return res.json as DiscoveryDoc;
|
|
60
|
+
} catch {
|
|
61
|
+
return null;
|
|
62
|
+
}
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
describe.skipIf(HTTP_SKIP)(
|
|
66
|
+
'prompt-mutation-workspace-membership-enforced: writes to non-member workspaces MUST be refused (RFC 0028 Tier-2)',
|
|
67
|
+
() => {
|
|
68
|
+
it('POST /v1/prompts with a workspaceId the principal is not a member of MUST NOT succeed with 2xx', async (ctx) => {
|
|
69
|
+
const d = await readDiscovery();
|
|
70
|
+
if (d === null) {
|
|
71
|
+
ctx.skip();
|
|
72
|
+
return;
|
|
73
|
+
}
|
|
74
|
+
const mutableLibrary = d.capabilities?.prompts?.mutableLibrary;
|
|
75
|
+
if (mutableLibrary !== true) {
|
|
76
|
+
ctx.skip();
|
|
77
|
+
return;
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
const nonMemberWorkspaceId =
|
|
81
|
+
process.env.OPENWOP_TEST_NONMEMBER_WORKSPACE_ID ??
|
|
82
|
+
`openwop-conformance-nonmember-${randomUUID()}`;
|
|
83
|
+
|
|
84
|
+
const res = await driver.post('/v1/prompts', {
|
|
85
|
+
workspaceId: nonMemberWorkspaceId,
|
|
86
|
+
templateId: `conformance-membership-probe-${randomUUID()}`,
|
|
87
|
+
version: '1.0.0',
|
|
88
|
+
kind: 'system',
|
|
89
|
+
text: 'conformance probe — SHOULD NOT persist',
|
|
90
|
+
});
|
|
91
|
+
|
|
92
|
+
// The conformance MUST: the host MUST NOT honor a write to a workspace
|
|
93
|
+
// the caller cannot prove membership of. Any refusal (4xx/5xx) is
|
|
94
|
+
// acceptable; a 2xx silent success is the failure mode that the RFC
|
|
95
|
+
// 0028 Tier-2 self-disclosed vulnerability demonstrated.
|
|
96
|
+
expect(
|
|
97
|
+
res.status,
|
|
98
|
+
driver.describe(
|
|
99
|
+
'spec/v1/prompts.md §Workspace membership on workspace-scoped reads and writes',
|
|
100
|
+
`mutating /v1/prompts MUST refuse a write to a non-member workspace; ` +
|
|
101
|
+
`got ${res.status} ${res.text.slice(0, 200)}`,
|
|
102
|
+
),
|
|
103
|
+
).toBeGreaterThanOrEqual(400);
|
|
104
|
+
|
|
105
|
+
// T1 canonicalization (2026-05-25): when the host CHOOSES 403 to
|
|
106
|
+
// signal the authz boundary, the response envelope MUST carry
|
|
107
|
+
// `error: "workspace_membership_required"` per rest-endpoints.md
|
|
108
|
+
// §"Common error codes". Hosts that refuse with other codes
|
|
109
|
+
// (401 if they treat the failure as authentication-level, 404 to
|
|
110
|
+
// avoid existence disclosure, 5xx on infra failure) have the
|
|
111
|
+
// refusal accepted above but the envelope shape is NOT constrained
|
|
112
|
+
// by this scenario — the canonical envelope is conditional on the
|
|
113
|
+
// 403 status code, not a forced upgrade.
|
|
114
|
+
if (res.status === 403) {
|
|
115
|
+
const body = res.json as { error?: unknown } | null;
|
|
116
|
+
expect(
|
|
117
|
+
body?.error,
|
|
118
|
+
driver.describe(
|
|
119
|
+
'spec/v1/rest-endpoints.md §Common error codes — workspace_membership_required',
|
|
120
|
+
`403 refusal of a workspace-scoped mutation MUST carry error: "workspace_membership_required"; got error: ${JSON.stringify(body?.error)}`,
|
|
121
|
+
),
|
|
122
|
+
).toBe('workspace_membership_required');
|
|
123
|
+
}
|
|
124
|
+
});
|
|
125
|
+
},
|
|
126
|
+
);
|
|
@@ -0,0 +1,183 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* prompt-read-workspace-membership-enforced — RFC 0028 Tier-2 §"Workspace
|
|
3
|
+
* membership on workspace-scoped reads and writes" verification (READ path).
|
|
4
|
+
*
|
|
5
|
+
* Status: ACTIVE (capability-gated; behavioral when the host advertises
|
|
6
|
+
* `capabilities.prompts.supported: true` AND accepts `?workspaceId=` on
|
|
7
|
+
* `GET /v1/prompts`). Hosts that don't expose workspace-scoped reads
|
|
8
|
+
* (host-only template libraries with no workspace dimension) self-skip
|
|
9
|
+
* via response-shape detection.
|
|
10
|
+
*
|
|
11
|
+
* The contract (spec/v1/prompts.md §"Discovery & distribution" §"REST
|
|
12
|
+
* endpoints" §"Workspace membership on workspace-scoped reads and writes"):
|
|
13
|
+
*
|
|
14
|
+
* Read paths are NOT exempt from the workspace-membership invariant
|
|
15
|
+
* just because they don't write. A GET /v1/prompts?workspaceId=<not-mine>
|
|
16
|
+
* that returns another workspace's templates is a cross-tenant data leak
|
|
17
|
+
* with the same blast radius as a cross-tenant write. Hosts MUST verify
|
|
18
|
+
* the authenticated principal's workspace membership BEFORE returning
|
|
19
|
+
* workspace-scoped content.
|
|
20
|
+
*
|
|
21
|
+
* Gate per MyndHyve relay 2026-05-25 ("Option B"): probe ALL hosts that
|
|
22
|
+
* advertise `capabilities.prompts.supported: true` regardless of
|
|
23
|
+
* `mutableLibrary`; read-only hosts that expose `?workspaceId=` reads are
|
|
24
|
+
* NOT exempt from the symmetric authz invariant. Hosts that don't expose
|
|
25
|
+
* workspace-scoped reads at all self-skip via the response interpretation
|
|
26
|
+
* below (the suite avoids inventing a new capability field just for this
|
|
27
|
+
* gating concern).
|
|
28
|
+
*
|
|
29
|
+
* The probe drives `GET /v1/prompts?workspaceId=<random-uuid>` and
|
|
30
|
+
* interprets the response:
|
|
31
|
+
*
|
|
32
|
+
* - 4xx (any code) — PASS (refused). If 403 specifically, additionally
|
|
33
|
+
* pin `error === "workspace_membership_required"` per the canonical
|
|
34
|
+
* envelope in rest-endpoints.md §"Common error codes".
|
|
35
|
+
* - 200 with `templates: []` — PASS. The host correctly returned no
|
|
36
|
+
* content for a workspace the principal isn't a member of. A random
|
|
37
|
+
* UUID workspace also definitionally has no real content, so an empty
|
|
38
|
+
* result is the correct null answer.
|
|
39
|
+
* - 200 with `templates: [non-empty]` — FAIL. The host returned content
|
|
40
|
+
* for an unauthorized workspace. This is the cross-tenant data leak
|
|
41
|
+
* failure mode. (Note: this scenario uses a random workspaceId so any
|
|
42
|
+
* non-empty result is a leak — there can't legitimately be templates
|
|
43
|
+
* in a freshly-generated nonexistent workspace.)
|
|
44
|
+
* - 200 without a `templates[]` field, or a response shape that doesn't
|
|
45
|
+
* resemble the documented `/v1/prompts` list shape — SKIP with a
|
|
46
|
+
* diagnostic log. Indicates the host doesn't recognize `?workspaceId=`
|
|
47
|
+
* on this endpoint (e.g., host-only template library with no
|
|
48
|
+
* workspace dimension).
|
|
49
|
+
* - 5xx — PASS (refused; envelope shape unconstrained).
|
|
50
|
+
*
|
|
51
|
+
* Why a random workspaceId is sufficient: the assertion is negative-space.
|
|
52
|
+
* A host that correctly enforces membership MUST refuse for ANY workspace
|
|
53
|
+
* the principal isn't a member of, and a random UUID has astronomically-low
|
|
54
|
+
* collision probability with any real workspace membership grant. A host
|
|
55
|
+
* that returns templates from a random UUID workspace is leaking content
|
|
56
|
+
* from somewhere (host-built-in misclassified as workspace, or a silent
|
|
57
|
+
* fall-through to another workspace's content, or a query bug returning
|
|
58
|
+
* everything).
|
|
59
|
+
*
|
|
60
|
+
* @see RFCS/0028-prompt-library-endpoints.md §"Post-promotion notes"
|
|
61
|
+
* @see spec/v1/prompts.md §"Security invariants" §prompt-read-workspace-membership-enforced
|
|
62
|
+
* @see spec/v1/rest-endpoints.md §"Common error codes" §workspace_membership_required
|
|
63
|
+
* @see spec/v1/auth.md §"Identity claims — tenant · workspace · principal"
|
|
64
|
+
* @see RFCS/0048-tenant-workspace-principal-identity-model.md §D
|
|
65
|
+
*/
|
|
66
|
+
|
|
67
|
+
import { describe, it, expect } from 'vitest';
|
|
68
|
+
import { randomUUID } from 'node:crypto';
|
|
69
|
+
import { driver } from '../lib/driver.js';
|
|
70
|
+
|
|
71
|
+
const HTTP_SKIP = !process.env.OPENWOP_BASE_URL;
|
|
72
|
+
|
|
73
|
+
interface DiscoveryDoc {
|
|
74
|
+
capabilities?: {
|
|
75
|
+
prompts?: {
|
|
76
|
+
supported?: unknown;
|
|
77
|
+
};
|
|
78
|
+
};
|
|
79
|
+
}
|
|
80
|
+
|
|
81
|
+
interface PromptListResponse {
|
|
82
|
+
templates?: unknown;
|
|
83
|
+
}
|
|
84
|
+
|
|
85
|
+
async function readDiscovery(): Promise<DiscoveryDoc | null> {
|
|
86
|
+
try {
|
|
87
|
+
const res = await driver.get('/.well-known/openwop');
|
|
88
|
+
if (res.status !== 200) return null;
|
|
89
|
+
return res.json as DiscoveryDoc;
|
|
90
|
+
} catch {
|
|
91
|
+
return null;
|
|
92
|
+
}
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
describe.skipIf(HTTP_SKIP)(
|
|
96
|
+
'prompt-read-workspace-membership-enforced: workspace-scoped reads MUST NOT leak templates from another workspace (RFC 0028 Tier-2)',
|
|
97
|
+
() => {
|
|
98
|
+
it('GET /v1/prompts?workspaceId=<non-member> MUST refuse OR return empty templates[] — never another workspace\'s content', async (ctx) => {
|
|
99
|
+
const d = await readDiscovery();
|
|
100
|
+
if (d === null) {
|
|
101
|
+
ctx.skip();
|
|
102
|
+
return;
|
|
103
|
+
}
|
|
104
|
+
const promptsSupported = d.capabilities?.prompts?.supported;
|
|
105
|
+
if (promptsSupported !== true) {
|
|
106
|
+
ctx.skip();
|
|
107
|
+
return;
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
const nonMemberWorkspaceId =
|
|
111
|
+
process.env.OPENWOP_TEST_NONMEMBER_WORKSPACE_ID ??
|
|
112
|
+
`openwop-conformance-nonmember-${randomUUID()}`;
|
|
113
|
+
|
|
114
|
+
const res = await driver.get(
|
|
115
|
+
`/v1/prompts?workspaceId=${encodeURIComponent(nonMemberWorkspaceId)}`,
|
|
116
|
+
);
|
|
117
|
+
|
|
118
|
+
// 4xx — refused. Acceptable shape for the membership-required failure
|
|
119
|
+
// (and any other refusal mode the host chooses: 401, 404 for
|
|
120
|
+
// existence-disclosure avoidance, etc).
|
|
121
|
+
if (res.status >= 400 && res.status < 500) {
|
|
122
|
+
// Canonical envelope on 403 per rest-endpoints.md §"Common error codes".
|
|
123
|
+
if (res.status === 403) {
|
|
124
|
+
const body = res.json as { error?: unknown } | null;
|
|
125
|
+
expect(
|
|
126
|
+
body?.error,
|
|
127
|
+
driver.describe(
|
|
128
|
+
'spec/v1/rest-endpoints.md §Common error codes — workspace_membership_required',
|
|
129
|
+
`403 refusal of a workspace-scoped read MUST carry error: "workspace_membership_required"; got error: ${JSON.stringify(body?.error)}`,
|
|
130
|
+
),
|
|
131
|
+
).toBe('workspace_membership_required');
|
|
132
|
+
}
|
|
133
|
+
return;
|
|
134
|
+
}
|
|
135
|
+
|
|
136
|
+
// 5xx — refused (infrastructure failure is acceptable; envelope shape
|
|
137
|
+
// unconstrained).
|
|
138
|
+
if (res.status >= 500) return;
|
|
139
|
+
|
|
140
|
+
// 2xx — must inspect the response body. The failure mode this
|
|
141
|
+
// invariant guards against is a 200 response that LEAKS templates
|
|
142
|
+
// from a workspace the principal isn't a member of.
|
|
143
|
+
if (res.status >= 200 && res.status < 300) {
|
|
144
|
+
const body = res.json as PromptListResponse | null;
|
|
145
|
+
if (
|
|
146
|
+
body === null ||
|
|
147
|
+
typeof body !== 'object' ||
|
|
148
|
+
!('templates' in body)
|
|
149
|
+
) {
|
|
150
|
+
// Host doesn't recognize `?workspaceId=` on this endpoint
|
|
151
|
+
// (response shape doesn't include the documented `templates[]`
|
|
152
|
+
// field). Soft-skip: this scenario probes hosts that expose
|
|
153
|
+
// workspace-scoped reads, and a host without that surface is
|
|
154
|
+
// simply out of scope.
|
|
155
|
+
ctx.skip();
|
|
156
|
+
return;
|
|
157
|
+
}
|
|
158
|
+
const templates = body.templates;
|
|
159
|
+
if (!Array.isArray(templates)) {
|
|
160
|
+
// Same: unrecognized shape, skip.
|
|
161
|
+
ctx.skip();
|
|
162
|
+
return;
|
|
163
|
+
}
|
|
164
|
+
|
|
165
|
+
// A random non-member workspaceId can never legitimately contain
|
|
166
|
+
// templates the caller is authorized to see. Any non-empty result
|
|
167
|
+
// is a cross-tenant data leak.
|
|
168
|
+
expect(
|
|
169
|
+
templates.length,
|
|
170
|
+
driver.describe(
|
|
171
|
+
'spec/v1/prompts.md §Workspace membership on workspace-scoped reads and writes',
|
|
172
|
+
`GET /v1/prompts?workspaceId=<random-non-member> MUST NOT return any templates; got ${templates.length} templates which is a cross-tenant data leak (the random workspaceId is freshly generated per probe and cannot legitimately contain authorized content)`,
|
|
173
|
+
),
|
|
174
|
+
).toBe(0);
|
|
175
|
+
return;
|
|
176
|
+
}
|
|
177
|
+
|
|
178
|
+
// Other status codes (1xx, 3xx) — soft-skip with note. Not a clear
|
|
179
|
+
// signal either way.
|
|
180
|
+
ctx.skip();
|
|
181
|
+
});
|
|
182
|
+
},
|
|
183
|
+
);
|
|
@@ -100,7 +100,10 @@ describe('redaction: /.well-known/openwop secrets+aiProviders shape contract', (
|
|
|
100
100
|
'when secrets.supported is true, scopes MUST be non-empty',
|
|
101
101
|
)).toBeGreaterThanOrEqual(1);
|
|
102
102
|
for (const scope of scopes) {
|
|
103
|
-
|
|
103
|
+
// Allowlist MUST track the `secrets.scopes` enum in capabilities.schema.json
|
|
104
|
+
// (`["tenant", "user", "run", "workspace"]`). `workspace` is the RFC 0046/0048
|
|
105
|
+
// sub-tenant scope — additive; hosts that advertise it (e.g. MyndHyve) are conformant.
|
|
106
|
+
expect(['tenant', 'user', 'run', 'workspace']).toContain(scope);
|
|
104
107
|
}
|
|
105
108
|
expect(s.resolution, driver.describe(
|
|
106
109
|
'capabilities.md §"Secrets"',
|