ai-shield-core 0.1.0 → 0.2.0
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/dist/audit/logger.d.ts.map +1 -1
- package/dist/audit/logger.js +13 -14
- package/dist/audit/types.js +1 -2
- package/dist/cache/lru.js +1 -5
- package/dist/canary/memory.d.ts +75 -0
- package/dist/canary/memory.d.ts.map +1 -0
- package/dist/canary/memory.js +194 -0
- package/dist/context/wrap-context.d.ts +105 -0
- package/dist/context/wrap-context.d.ts.map +1 -0
- package/dist/context/wrap-context.js +188 -0
- package/dist/cost/anomaly.js +1 -4
- package/dist/cost/pricing.d.ts.map +1 -1
- package/dist/cost/pricing.js +18 -19
- package/dist/cost/tracker.d.ts +19 -1
- package/dist/cost/tracker.d.ts.map +1 -1
- package/dist/cost/tracker.js +27 -10
- package/dist/index.d.ts +31 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +51 -37
- package/dist/policy/circuit-breaker.d.ts +70 -0
- package/dist/policy/circuit-breaker.d.ts.map +1 -0
- package/dist/policy/circuit-breaker.js +376 -0
- package/dist/policy/engine.js +1 -5
- package/dist/policy/tools.js +4 -8
- package/dist/scanner/canary.js +4 -8
- package/dist/scanner/chain.js +1 -5
- package/dist/scanner/heuristic.d.ts +13 -0
- package/dist/scanner/heuristic.d.ts.map +1 -1
- package/dist/scanner/heuristic.js +50 -7
- package/dist/scanner/ingestion.d.ts +116 -0
- package/dist/scanner/ingestion.d.ts.map +1 -0
- package/dist/scanner/ingestion.js +452 -0
- package/dist/scanner/pii.d.ts.map +1 -1
- package/dist/scanner/pii.js +24 -12
- package/dist/shield.d.ts.map +1 -1
- package/dist/shield.js +34 -26
- package/dist/types.d.ts +140 -2
- package/dist/types.d.ts.map +1 -1
- package/dist/types.js +1 -2
- package/package.json +4 -3
- package/src/audit/logger.ts +6 -1
- package/src/canary/memory.ts +259 -0
- package/src/context/wrap-context.ts +304 -0
- package/src/cost/pricing.ts +13 -9
- package/src/cost/tracker.ts +35 -1
- package/src/index.ts +82 -1
- package/src/policy/circuit-breaker.ts +449 -0
- package/src/scanner/heuristic.ts +49 -2
- package/src/scanner/ingestion.ts +550 -0
- package/src/scanner/pii.ts +21 -7
- package/src/shield.ts +15 -2
- package/src/types.ts +175 -2
- package/tsconfig.json +2 -1
- package/dist/audit/logger.js.map +0 -1
- package/dist/audit/types.js.map +0 -1
- package/dist/cache/lru.js.map +0 -1
- package/dist/cost/anomaly.js.map +0 -1
- package/dist/cost/pricing.js.map +0 -1
- package/dist/cost/tracker.js.map +0 -1
- package/dist/index.js.map +0 -1
- package/dist/policy/engine.js.map +0 -1
- package/dist/policy/tools.js.map +0 -1
- package/dist/scanner/canary.js.map +0 -1
- package/dist/scanner/chain.js.map +0 -1
- package/dist/scanner/heuristic.js.map +0 -1
- package/dist/scanner/pii.js.map +0 -1
- package/dist/shield.js.map +0 -1
- package/dist/types.js.map +0 -1
package/dist/shield.js
CHANGED
|
@@ -1,18 +1,15 @@
|
|
|
1
|
-
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
7
|
-
|
|
8
|
-
|
|
9
|
-
const tracker_js_1 = require("./cost/tracker.js");
|
|
10
|
-
const logger_js_1 = require("./audit/logger.js");
|
|
11
|
-
const lru_js_1 = require("./cache/lru.js");
|
|
1
|
+
import { ScannerChain } from "./scanner/chain.js";
|
|
2
|
+
import { HeuristicScanner } from "./scanner/heuristic.js";
|
|
3
|
+
import { PIIScanner } from "./scanner/pii.js";
|
|
4
|
+
import { ToolPolicyScanner } from "./policy/tools.js";
|
|
5
|
+
import { PolicyEngine } from "./policy/engine.js";
|
|
6
|
+
import { CostTracker } from "./cost/tracker.js";
|
|
7
|
+
import { AuditLogger, ConsoleAuditStore } from "./audit/logger.js";
|
|
8
|
+
import { ScanLRUCache } from "./cache/lru.js";
|
|
12
9
|
// ============================================================
|
|
13
10
|
// AIShield — Main class, single entry point
|
|
14
11
|
// ============================================================
|
|
15
|
-
class AIShield {
|
|
12
|
+
export class AIShield {
|
|
16
13
|
chain;
|
|
17
14
|
policyEngine;
|
|
18
15
|
costTracker;
|
|
@@ -21,19 +18,19 @@ class AIShield {
|
|
|
21
18
|
config;
|
|
22
19
|
constructor(config = {}) {
|
|
23
20
|
this.config = config;
|
|
24
|
-
this.policyEngine = new
|
|
25
|
-
this.chain = new
|
|
21
|
+
this.policyEngine = new PolicyEngine(config.preset ?? "public_website");
|
|
22
|
+
this.chain = new ScannerChain({ earlyExit: true });
|
|
26
23
|
// Build scanner chain based on config
|
|
27
24
|
this.setupScanners(config);
|
|
28
25
|
// Cost tracker (optional, needs Redis for distributed use)
|
|
29
26
|
this.costTracker = config.cost?.enabled !== false && config.cost?.budgets
|
|
30
|
-
? new
|
|
27
|
+
? new CostTracker(config.cost.budgets)
|
|
31
28
|
: null;
|
|
32
29
|
// Audit logger (optional)
|
|
33
30
|
this.auditLogger = this.setupAudit(config);
|
|
34
31
|
// Scan cache (enabled when cache config is provided)
|
|
35
32
|
this.scanCache = config.cache && config.cache.enabled !== false
|
|
36
|
-
? new
|
|
33
|
+
? new ScanLRUCache({
|
|
37
34
|
maxSize: config.cache.maxSize,
|
|
38
35
|
ttlMs: config.cache.ttlMs,
|
|
39
36
|
})
|
|
@@ -41,6 +38,15 @@ class AIShield {
|
|
|
41
38
|
}
|
|
42
39
|
/** Scan input text — the main API */
|
|
43
40
|
async scan(input, context = {}) {
|
|
41
|
+
// Input-length guard — without this a user-supplied multi-MB prompt would
|
|
42
|
+
// trigger every regex scan on the full buffer, which is O(n) best-case
|
|
43
|
+
// but pathological under ReDoS-prone patterns. 256 KB handles every
|
|
44
|
+
// real chat/tool input we care about with plenty of headroom; override
|
|
45
|
+
// via AI_SHIELD_MAX_INPUT_BYTES when needed.
|
|
46
|
+
const maxInputBytes = Number(process.env.AI_SHIELD_MAX_INPUT_BYTES ?? 262_144);
|
|
47
|
+
if (input.length > maxInputBytes) {
|
|
48
|
+
input = input.slice(0, maxInputBytes);
|
|
49
|
+
}
|
|
44
50
|
// Apply preset if not set in context
|
|
45
51
|
if (!context.preset) {
|
|
46
52
|
context.preset = this.config.preset ?? "public_website";
|
|
@@ -59,8 +65,11 @@ class AIShield {
|
|
|
59
65
|
const cacheKey = this.buildCacheKey(input, context);
|
|
60
66
|
this.scanCache.set(cacheKey, result);
|
|
61
67
|
}
|
|
62
|
-
// Log to audit if enabled
|
|
63
|
-
|
|
68
|
+
// Log to audit if enabled — but never double-log the same input when
|
|
69
|
+
// a downstream caller re-scans cached content (result.meta.cached is set
|
|
70
|
+
// by the cache hit path above; here it is always false but we guard to
|
|
71
|
+
// be explicit and so subclasses extending scan() stay safe).
|
|
72
|
+
if (this.auditLogger && !result.meta.cached) {
|
|
64
73
|
void this.auditLogger.log(input, result, context);
|
|
65
74
|
}
|
|
66
75
|
return result;
|
|
@@ -117,7 +126,7 @@ class AIShield {
|
|
|
117
126
|
// 1. Heuristic injection scanner (always on unless explicitly disabled)
|
|
118
127
|
if (config.injection?.enabled !== false) {
|
|
119
128
|
const preset = this.policyEngine.getPreset();
|
|
120
|
-
this.chain.add(new
|
|
129
|
+
this.chain.add(new HeuristicScanner({
|
|
121
130
|
strictness: config.injection?.strictness ?? "medium",
|
|
122
131
|
threshold: config.injection?.threshold ?? preset.injection.threshold,
|
|
123
132
|
customPatterns: config.injection?.customPatterns?.map((pattern, i) => ({
|
|
@@ -131,7 +140,7 @@ class AIShield {
|
|
|
131
140
|
}
|
|
132
141
|
// 2. PII scanner
|
|
133
142
|
if (config.pii?.enabled !== false) {
|
|
134
|
-
this.chain.add(new
|
|
143
|
+
this.chain.add(new PIIScanner({
|
|
135
144
|
action: config.pii?.action ?? this.policyEngine.getPIIAction(),
|
|
136
145
|
locale: config.pii?.locale,
|
|
137
146
|
types: config.pii?.types,
|
|
@@ -149,7 +158,7 @@ class AIShield {
|
|
|
149
158
|
this.policyEngine.getMaxToolChainDepth(),
|
|
150
159
|
},
|
|
151
160
|
};
|
|
152
|
-
this.chain.add(new
|
|
161
|
+
this.chain.add(new ToolPolicyScanner(toolPolicy, config.tools.manifestPins));
|
|
153
162
|
}
|
|
154
163
|
}
|
|
155
164
|
setupAudit(config) {
|
|
@@ -158,27 +167,26 @@ class AIShield {
|
|
|
158
167
|
let store;
|
|
159
168
|
switch (config.audit?.store) {
|
|
160
169
|
case "console":
|
|
161
|
-
store = new
|
|
170
|
+
store = new ConsoleAuditStore();
|
|
162
171
|
break;
|
|
163
172
|
case "postgresql":
|
|
164
173
|
// PostgreSQL store would be imported separately to keep core lightweight
|
|
165
174
|
// For now, fall through to console
|
|
166
|
-
store = new
|
|
175
|
+
store = new ConsoleAuditStore();
|
|
167
176
|
break;
|
|
168
177
|
case "memory":
|
|
169
178
|
default:
|
|
170
179
|
// If no store configured and audit not explicitly enabled, skip
|
|
171
180
|
if (!config.audit?.store && config.audit?.enabled !== true)
|
|
172
181
|
return null;
|
|
173
|
-
store = new
|
|
182
|
+
store = new ConsoleAuditStore();
|
|
174
183
|
break;
|
|
175
184
|
}
|
|
176
|
-
return new
|
|
185
|
+
return new AuditLogger({
|
|
177
186
|
store,
|
|
178
187
|
batchSize: config.audit?.batchSize,
|
|
179
188
|
flushIntervalMs: config.audit?.flushIntervalMs,
|
|
180
189
|
});
|
|
181
190
|
}
|
|
182
191
|
}
|
|
183
|
-
exports.AIShield = AIShield;
|
|
184
192
|
//# sourceMappingURL=shield.js.map
|
package/dist/types.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
export type ScanDecision = "allow" | "warn" | "block";
|
|
2
|
-
export type ViolationType = "prompt_injection" | "pii_detected" | "tool_denied" | "tool_rate_limit" | "budget_exceeded" | "content_policy" | "manifest_drift";
|
|
2
|
+
export type ViolationType = "prompt_injection" | "pii_detected" | "tool_denied" | "tool_rate_limit" | "budget_exceeded" | "content_policy" | "manifest_drift" | "ingested_injection" | "untrusted_instruction" | "memory_poisoning" | "circuit_breaker_open" | "blast_radius_exceeded";
|
|
3
3
|
export interface Violation {
|
|
4
4
|
type: ViolationType;
|
|
5
5
|
scanner: string;
|
|
@@ -29,16 +29,154 @@ export interface Scanner {
|
|
|
29
29
|
name: string;
|
|
30
30
|
scan(input: string, context: ScanContext): Promise<ScannerResult>;
|
|
31
31
|
}
|
|
32
|
+
/**
|
|
33
|
+
* Where a piece of content came from. Determines how aggressively the
|
|
34
|
+
* ingestion scanner treats instruction-like patterns.
|
|
35
|
+
*
|
|
36
|
+
* - `user` — Direct user message. Treat normally.
|
|
37
|
+
* - `rag` — Retrieved document chunk (vector store, knowledge base).
|
|
38
|
+
* Instructions here are almost always an indirect-injection
|
|
39
|
+
* attempt.
|
|
40
|
+
* - `tool-desc` — MCP tool description / OpenAI function schema / tool args
|
|
41
|
+
* that came from a remote MCP server. High-risk vector
|
|
42
|
+
* per Lakera 2026 advisory + OX Security MCP CVEs.
|
|
43
|
+
* - `memory` — Persisted memory entry (knowledge graph, session
|
|
44
|
+
* history, vector memory). Subject to persistence-poisoning.
|
|
45
|
+
* - `web` — Scraped web page / HTML. Hidden-instruction risk via
|
|
46
|
+
* HTML comments, CSS-hidden text, unicode tricks.
|
|
47
|
+
* - `agent-output`— Output from another agent flowing into this one
|
|
48
|
+
* (multi-agent contagion).
|
|
49
|
+
*/
|
|
50
|
+
export type IngestionSource = "user" | "rag" | "tool-desc" | "memory" | "web" | "agent-output";
|
|
51
|
+
/**
|
|
52
|
+
* Privilege tier of a content segment. The toolkit treats instructions
|
|
53
|
+
* coming from `untrusted` sources differently from those in `system`
|
|
54
|
+
* or `trusted` segments. See `wrapContext()`.
|
|
55
|
+
*/
|
|
56
|
+
export type TrustTier = "system" | "trusted" | "untrusted";
|
|
32
57
|
export interface ScanContext {
|
|
33
58
|
agentId?: string;
|
|
34
59
|
sessionId?: string;
|
|
35
60
|
userId?: string;
|
|
36
|
-
userType?:
|
|
61
|
+
userType?: string;
|
|
37
62
|
locale?: string;
|
|
38
63
|
preset?: PresetName;
|
|
39
64
|
tools?: ToolCall[];
|
|
65
|
+
/**
|
|
66
|
+
* Provenance of the content being scanned. Defaults to `"user"` when
|
|
67
|
+
* unset. Set to `"rag"`/`"tool-desc"`/`"memory"`/`"web"`/`"agent-output"`
|
|
68
|
+
* to engage indirect-injection heuristics.
|
|
69
|
+
*/
|
|
70
|
+
source?: IngestionSource;
|
|
71
|
+
/**
|
|
72
|
+
* Privilege tier of the content. Defaults inferred from `source`:
|
|
73
|
+
* `user` → untrusted, everything else → untrusted, `system` only when
|
|
74
|
+
* explicitly set via `wrapContext()`.
|
|
75
|
+
*/
|
|
76
|
+
trustTier?: TrustTier;
|
|
40
77
|
}
|
|
41
78
|
export type PresetName = "public_website" | "internal_support" | "ops_agent";
|
|
79
|
+
/**
|
|
80
|
+
* A single content segment with provenance + trust tagging.
|
|
81
|
+
* Produced by `wrapContext()`, consumed by `assemblePrompt()` and
|
|
82
|
+
* any policy that needs to enforce tier boundaries.
|
|
83
|
+
*/
|
|
84
|
+
export interface ContextSegment {
|
|
85
|
+
/** Source of the segment. */
|
|
86
|
+
source: IngestionSource;
|
|
87
|
+
/** Privilege tier. */
|
|
88
|
+
trust: TrustTier;
|
|
89
|
+
/** Raw text content. */
|
|
90
|
+
content: string;
|
|
91
|
+
/** Optional human-readable origin label (e.g. "wikipedia.org/wiki/X"). */
|
|
92
|
+
label?: string;
|
|
93
|
+
/** Optional SHA-256 content hash for poisoning detection. */
|
|
94
|
+
contentHash?: string;
|
|
95
|
+
}
|
|
96
|
+
export interface WrappedContext {
|
|
97
|
+
/** All segments in the order they were added. */
|
|
98
|
+
segments: ContextSegment[];
|
|
99
|
+
/** Per-segment scan results (filled by `scanWrappedContext()`). */
|
|
100
|
+
scanResults?: Array<{
|
|
101
|
+
segmentIndex: number;
|
|
102
|
+
decision: ScanDecision;
|
|
103
|
+
violations: Violation[];
|
|
104
|
+
}>;
|
|
105
|
+
/** Aggregate decision across all segments. */
|
|
106
|
+
decision?: ScanDecision;
|
|
107
|
+
}
|
|
108
|
+
/**
|
|
109
|
+
* A memory entry sealed with a content hash + sentinel canary.
|
|
110
|
+
* `verifyMemoryCanary()` re-derives the hash on read and flags
|
|
111
|
+
* silent mutation.
|
|
112
|
+
*/
|
|
113
|
+
export interface MemoryCanaryEntry {
|
|
114
|
+
/** Stable identifier of the memory entry (e.g. "fact:12345"). */
|
|
115
|
+
id: string;
|
|
116
|
+
/** Original content at write time. */
|
|
117
|
+
content: string;
|
|
118
|
+
/** SHA-256(`${id}\0${content}\0${canaryToken}`). */
|
|
119
|
+
contentHash: string;
|
|
120
|
+
/** Random sentinel token. Used to bind verification to this write. */
|
|
121
|
+
canaryToken: string;
|
|
122
|
+
/** When the canary was minted. */
|
|
123
|
+
createdAt: Date;
|
|
124
|
+
/** Optional tenant scope so leaks across tenants are detectable. */
|
|
125
|
+
tenantId?: string;
|
|
126
|
+
}
|
|
127
|
+
export interface MemoryCanaryVerification {
|
|
128
|
+
valid: boolean;
|
|
129
|
+
reason?: "content_mutated" | "canary_missing" | "tenant_mismatch" | "hash_mismatch";
|
|
130
|
+
/** When invalid: the mutated content actually read. */
|
|
131
|
+
observed?: string;
|
|
132
|
+
}
|
|
133
|
+
export type CircuitState = "closed" | "open" | "half-open";
|
|
134
|
+
export interface CircuitBreakerConfig {
|
|
135
|
+
/** Logical tool name this breaker applies to. */
|
|
136
|
+
tool: string;
|
|
137
|
+
/** Optional sub-scope (per-agent, per-session, per-tenant). */
|
|
138
|
+
scope?: string;
|
|
139
|
+
/** Anomaly count that trips the breaker. Default: 5. */
|
|
140
|
+
failureThreshold?: number;
|
|
141
|
+
/** Rolling window in ms for the failure count. Default: 60_000. */
|
|
142
|
+
windowMs?: number;
|
|
143
|
+
/** How long the breaker stays open after tripping (ms). Default: 60_000. */
|
|
144
|
+
cooldownMs?: number;
|
|
145
|
+
/** Max calls per `windowMs`. Default: unlimited. */
|
|
146
|
+
maxCallsPerWindow?: number;
|
|
147
|
+
/** Max "destructive" calls per `windowMs`. Default: unlimited. */
|
|
148
|
+
maxWritesPerWindow?: number;
|
|
149
|
+
/** Optional human-in-the-loop confirmation hook. */
|
|
150
|
+
onDestructive?: (info: {
|
|
151
|
+
tool: string;
|
|
152
|
+
scope?: string;
|
|
153
|
+
context: ScanContext;
|
|
154
|
+
}) => boolean | Promise<boolean>;
|
|
155
|
+
/** Treat this tool as destructive (default: inferred via wildcard list). */
|
|
156
|
+
isDestructive?: boolean;
|
|
157
|
+
}
|
|
158
|
+
export interface CircuitBreakerDecision {
|
|
159
|
+
/** Whether the call may proceed. */
|
|
160
|
+
allowed: boolean;
|
|
161
|
+
/** Current state. */
|
|
162
|
+
state: CircuitState;
|
|
163
|
+
/** Reason if `allowed = false`. */
|
|
164
|
+
reason?: "circuit_open" | "rate_limit" | "blast_radius_exceeded" | "hitl_denied";
|
|
165
|
+
/** Detail for logs. */
|
|
166
|
+
message?: string;
|
|
167
|
+
/** Suggested retry-after in ms when state = open. */
|
|
168
|
+
retryAfterMs?: number;
|
|
169
|
+
}
|
|
170
|
+
/**
|
|
171
|
+
* Minimal counter store usable by the circuit breaker.
|
|
172
|
+
* Compatible subset of `ioredis` so existing
|
|
173
|
+
* `cost.tracker.RedisLike` deployments can be reused.
|
|
174
|
+
*/
|
|
175
|
+
export interface CounterStoreLike {
|
|
176
|
+
get(key: string): Promise<string | null>;
|
|
177
|
+
incrbyfloat(key: string, increment: number): Promise<string>;
|
|
178
|
+
expire(key: string, seconds: number): Promise<number>;
|
|
179
|
+
}
|
|
42
180
|
export type PIIType = "email" | "phone" | "iban" | "credit_card" | "german_tax_id" | "german_personal_id" | "german_social_security" | "ip_address" | "url_with_credentials";
|
|
43
181
|
export type PIIAction = "block" | "mask" | "tokenize" | "allow";
|
|
44
182
|
export interface PIIEntity {
|
package/dist/types.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAMA,MAAM,MAAM,YAAY,GAAG,OAAO,GAAG,MAAM,GAAG,OAAO,CAAC;AAEtD,MAAM,MAAM,aAAa,GACrB,kBAAkB,GAClB,cAAc,GACd,aAAa,GACb,iBAAiB,GACjB,iBAAiB,GACjB,gBAAgB,GAChB,gBAAgB,CAAC;
|
|
1
|
+
{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAMA,MAAM,MAAM,YAAY,GAAG,OAAO,GAAG,MAAM,GAAG,OAAO,CAAC;AAEtD,MAAM,MAAM,aAAa,GACrB,kBAAkB,GAClB,cAAc,GACd,aAAa,GACb,iBAAiB,GACjB,iBAAiB,GACjB,gBAAgB,GAChB,gBAAgB,GAChB,oBAAoB,GACpB,uBAAuB,GACvB,kBAAkB,GAClB,sBAAsB,GACtB,uBAAuB,CAAC;AAE5B,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,aAAa,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,UAAU;IACzB,IAAI,EAAE,OAAO,CAAC;IACd,QAAQ,EAAE,YAAY,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB,IAAI,EAAE;QACJ,cAAc,EAAE,MAAM,CAAC;QACvB,WAAW,EAAE,MAAM,EAAE,CAAC;QACtB,MAAM,EAAE,OAAO,CAAC;KACjB,CAAC;CACH;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,YAAY,CAAC;IACvB,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,WAAW,GAAG,OAAO,CAAC,aAAa,CAAC,CAAC;CACnE;AAID;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,MAAM,eAAe,GACvB,MAAM,GACN,KAAK,GACL,WAAW,GACX,QAAQ,GACR,KAAK,GACL,cAAc,CAAC;AAEnB;;;;GAIG;AACH,MAAM,MAAM,SAAS,GAAG,QAAQ,GAAG,SAAS,GAAG,WAAW,CAAC;AAE3D,MAAM,WAAW,WAAW;IAC1B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,UAAU,CAAC;IACpB,KAAK,CAAC,EAAE,QAAQ,EAAE,CAAC;IACnB;;;;OAIG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB;;;;OAIG;IACH,SAAS,CAAC,EAAE,SAAS,CAAC;CACvB;AAED,MAAM,MAAM,UAAU,GAAG,gBAAgB,GAAG,kBAAkB,GAAG,WAAW,CAAC;AAI7E;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,6BAA6B;IAC7B,MAAM,EAAE,eAAe,CAAC;IACxB,sBAAsB;IACtB,KAAK,EAAE,SAAS,CAAC;IACjB,wBAAwB;IACxB,OAAO,EAAE,MAAM,CAAC;IAChB,0EAA0E;IAC1E,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,6DAA6D;IAC7D,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,cAAc;IAC7B,iDAAiD;IACjD,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,mEAAmE;IACnE,WAAW,CAAC,EAAE,KAAK,CAAC;QAClB,YAAY,EAAE,MAAM,CAAC;QACrB,QAAQ,EAAE,YAAY,CAAC;QACvB,UAAU,EAAE,SAAS,EAAE,CAAC;KACzB,CAAC,CAAC;IACH,8CAA8C;IAC9C,QAAQ,CAAC,EAAE,YAAY,CAAC;CACzB;AAID;;;;GAIG;AACH,MAAM,WAAW,iBAAiB;IAChC,iEAAiE;IACjE,EAAE,EAAE,MAAM,CAAC;IACX,sCAAsC;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,oDAAoD;IACpD,WAAW,EAAE,MAAM,CAAC;IACpB,sEAAsE;IACtE,WAAW,EAAE,MAAM,CAAC;IACpB,kCAAkC;IAClC,SAAS,EAAE,IAAI,CAAC;IAChB,oEAAoE;IACpE,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,wBAAwB;IACvC,KAAK,EAAE,OAAO,CAAC;IACf,MAAM,CAAC,EACH,iBAAiB,GACjB,gBAAgB,GAChB,iBAAiB,GACjB,eAAe,CAAC;IACpB,uDAAuD;IACvD,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAID,MAAM,MAAM,YAAY,GAAG,QAAQ,GAAG,MAAM,GAAG,WAAW,CAAC;AAE3D,MAAM,WAAW,oBAAoB;IACnC,iDAAiD;IACjD,IAAI,EAAE,MAAM,CAAC;IACb,+DAA+D;IAC/D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,wDAAwD;IACxD,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,mEAAmE;IACnE,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,4EAA4E;IAC5E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oDAAoD;IACpD,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,kEAAkE;IAClE,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,oDAAoD;IACpD,aAAa,CAAC,EAAE,CAAC,IAAI,EAAE;QACrB,IAAI,EAAE,MAAM,CAAC;QACb,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,OAAO,EAAE,WAAW,CAAC;KACtB,KAAK,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IACjC,4EAA4E;IAC5E,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AAED,MAAM,WAAW,sBAAsB;IACrC,oCAAoC;IACpC,OAAO,EAAE,OAAO,CAAC;IACjB,qBAAqB;IACrB,KAAK,EAAE,YAAY,CAAC;IACpB,mCAAmC;IACnC,MAAM,CAAC,EACH,cAAc,GACd,YAAY,GACZ,uBAAuB,GACvB,aAAa,CAAC;IAClB,uBAAuB;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,qDAAqD;IACrD,YAAY,CAAC,EAAE,MAAM,CAAC;CACvB;AAED;;;;GAIG;AACH,MAAM,WAAW,gBAAgB;IAC/B,GAAG,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;IACzC,WAAW,CAAC,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;IAC7D,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;CACvD;AAID,MAAM,MAAM,OAAO,GACf,OAAO,GACP,OAAO,GACP,MAAM,GACN,aAAa,GACb,eAAe,GACf,oBAAoB,GACpB,wBAAwB,GACxB,YAAY,GACZ,sBAAsB,CAAC;AAE3B,MAAM,MAAM,SAAS,GAAG,OAAO,GAAG,MAAM,GAAG,UAAU,GAAG,OAAO,CAAC;AAEhE,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,OAAO,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,EAAE,MAAM,CAAC;CACpB;AAID,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACpC,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,EAAE,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,UAAU;IACzB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC7C,MAAM,CAAC,EAAE;QACP,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;QAC7B,YAAY,CAAC,EAAE,OAAO,CAAC;QACvB,iBAAiB,CAAC,EAAE,MAAM,CAAC;KAC5B,CAAC;CACH;AAED,MAAM,WAAW,eAAe;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,QAAQ,EAAE,IAAI,CAAC;CAChB;AAID,MAAM,MAAM,YAAY,GAAG,QAAQ,GAAG,OAAO,GAAG,SAAS,CAAC;AAE1D,MAAM,WAAW,YAAY;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,YAAY,CAAC;CACtB;AAED,MAAM,WAAW,YAAY;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,UAAU;IACzB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,IAAI,CAAC;CACjB;AAED,MAAM,WAAW,iBAAiB;IAChC,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAID,MAAM,WAAW,WAAW;IAC1B,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,IAAI,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,GAAG,WAAW,GAAG,gBAAgB,CAAC;IACrD,SAAS,EAAE,MAAM,CAAC;IAClB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,gBAAgB,EAAE,YAAY,CAAC;IAC/B,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB,cAAc,EAAE,MAAM,CAAC;IACvB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;IACvB,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB;AAID,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,UAAU,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,CAAC;IACvC,MAAM,CAAC,EAAE,OAAO,GAAG,MAAM,GAAG,MAAM,CAAC;IACnC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;CAC3B;AAED,MAAM,WAAW,SAAS;IACxB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,MAAM,CAAC,EAAE,SAAS,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC,CAAC;IAC5C,YAAY,CAAC,EAAE,OAAO,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACvC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtE,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,YAAY,GAAG,QAAQ,GAAG,SAAS,CAAC;IAC5C,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,UAAU;IACzB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC3C,uBAAuB,CAAC,EAAE,MAAM,EAAE,CAAC;IACnC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,YAAY,CAAC,EAAE,eAAe,EAAE,CAAC;CAClC;AAED,MAAM,WAAW,WAAW;IAC1B,uEAAuE;IACvE,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,6CAA6C;IAC7C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,yDAAyD;IACzD,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,YAAY;IAC3B,SAAS,CAAC,EAAE,eAAe,CAAC;IAC5B,GAAG,CAAC,EAAE,SAAS,CAAC;IAChB,IAAI,CAAC,EAAE,UAAU,CAAC;IAClB,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,KAAK,CAAC,EAAE,UAAU,CAAC;IACnB,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,MAAM,CAAC,EAAE,UAAU,CAAC;CACrB;AAID,MAAM,WAAW,YAAY;IAC3B,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B"}
|
package/dist/types.js
CHANGED
|
@@ -1,6 +1,5 @@
|
|
|
1
|
-
"use strict";
|
|
2
1
|
// ============================================================
|
|
3
2
|
// AI Shield Core Types
|
|
4
3
|
// ============================================================
|
|
5
|
-
|
|
4
|
+
export {};
|
|
6
5
|
//# sourceMappingURL=types.js.map
|
package/package.json
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "ai-shield-core",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.2.0",
|
|
4
|
+
"type": "module",
|
|
4
5
|
"description": "LLM Security SDK — Prompt Injection Detection, PII Protection, Cost Control, Audit",
|
|
5
6
|
"main": "dist/index.js",
|
|
6
7
|
"types": "dist/index.d.ts",
|
|
@@ -11,8 +12,8 @@
|
|
|
11
12
|
}
|
|
12
13
|
},
|
|
13
14
|
"scripts": {
|
|
14
|
-
"build": "tsc",
|
|
15
|
-
"typecheck": "tsc
|
|
15
|
+
"build": "tsc -b",
|
|
16
|
+
"typecheck": "tsc -b"
|
|
16
17
|
},
|
|
17
18
|
"dependencies": {},
|
|
18
19
|
"peerDependencies": {
|
package/src/audit/logger.ts
CHANGED
|
@@ -28,6 +28,11 @@ export class AuditLogger {
|
|
|
28
28
|
this.flushTimer = setInterval(() => {
|
|
29
29
|
void this.flush();
|
|
30
30
|
}, flushMs);
|
|
31
|
+
// Allow the Node process to exit even if the timer is pending —
|
|
32
|
+
// clean shutdown should go through close(), not rely on the timer.
|
|
33
|
+
if (typeof this.flushTimer.unref === "function") {
|
|
34
|
+
this.flushTimer.unref();
|
|
35
|
+
}
|
|
31
36
|
}
|
|
32
37
|
|
|
33
38
|
/** Log a scan result */
|
|
@@ -48,7 +53,7 @@ export class AuditLogger {
|
|
|
48
53
|
sessionId: context.sessionId,
|
|
49
54
|
agentId: context.agentId,
|
|
50
55
|
userIdHash: context.userId
|
|
51
|
-
? createHash("sha256").update(context.userId).digest("hex").substring(0,
|
|
56
|
+
? createHash("sha256").update(context.userId).digest("hex").substring(0, 32)
|
|
52
57
|
: undefined,
|
|
53
58
|
requestType: context.tools?.length ? "tool_call" : "chat",
|
|
54
59
|
inputHash: createHash("sha256").update(input).digest("hex"),
|
|
@@ -0,0 +1,259 @@
|
|
|
1
|
+
import { randomBytes, createHash, timingSafeEqual } from "node:crypto";
|
|
2
|
+
import type {
|
|
3
|
+
MemoryCanaryEntry,
|
|
4
|
+
MemoryCanaryVerification,
|
|
5
|
+
} from "../types.js";
|
|
6
|
+
|
|
7
|
+
// ============================================================
|
|
8
|
+
// Memory Canary — Persistence-Poisoning Detection
|
|
9
|
+
//
|
|
10
|
+
// Existing canary tokens defend the *system prompt*: inject a sentinel
|
|
11
|
+
// + check the response for leakage. That works for prompt-extraction
|
|
12
|
+
// attacks but does nothing for the broader class of *persistence
|
|
13
|
+
// poisoning*: an attacker mutates a stored memory entry, knowledge-graph
|
|
14
|
+
// fact, or RAG document so that the next retrieval steers the model.
|
|
15
|
+
//
|
|
16
|
+
// MemoryCanary seals each write with:
|
|
17
|
+
// 1. A random sentinel token bound to the entry (`canaryToken`).
|
|
18
|
+
// 2. A SHA-256 hash over `id || content || canaryToken` so any silent
|
|
19
|
+
// mutation of `content` between write and read is detectable.
|
|
20
|
+
// 3. Optional `tenantId` binding so a cross-tenant leak surfaces as
|
|
21
|
+
// an explicit `tenant_mismatch` reason.
|
|
22
|
+
//
|
|
23
|
+
// Storage of the canary metadata is the caller's responsibility — the
|
|
24
|
+
// returned `MemoryCanaryEntry` is JSON-serialisable and meant to be
|
|
25
|
+
// persisted alongside the memory entry (separate column / sidecar key).
|
|
26
|
+
//
|
|
27
|
+
// The library does not store secrets and does not need a key. The
|
|
28
|
+
// security property is integrity (mutation detection), not
|
|
29
|
+
// confidentiality.
|
|
30
|
+
// ============================================================
|
|
31
|
+
|
|
32
|
+
/** Minimum allowed canary-token length in bytes (= 16 hex chars). */
|
|
33
|
+
const MIN_TOKEN_BYTES = 8;
|
|
34
|
+
/** Default canary-token length: 16 random bytes -> 32 hex chars. */
|
|
35
|
+
const DEFAULT_TOKEN_BYTES = 16;
|
|
36
|
+
|
|
37
|
+
export interface MintMemoryCanaryOptions {
|
|
38
|
+
/**
|
|
39
|
+
* Override token byte length. Minimum 8 (16 hex chars). Default 16.
|
|
40
|
+
* Longer = stronger guessing resistance, but the hex string lives in
|
|
41
|
+
* sidecar storage so the overhead is negligible.
|
|
42
|
+
*/
|
|
43
|
+
tokenBytes?: number;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
/**
|
|
47
|
+
* Mint a canary for a memory entry. Call at write time, persist the
|
|
48
|
+
* returned `MemoryCanaryEntry` alongside (or in place of) the raw entry.
|
|
49
|
+
*
|
|
50
|
+
* @param id Stable identifier of the memory entry.
|
|
51
|
+
* @param content The content being stored.
|
|
52
|
+
* @param tenantId Optional tenant scope.
|
|
53
|
+
*/
|
|
54
|
+
export function mintMemoryCanary(
|
|
55
|
+
id: string,
|
|
56
|
+
content: string,
|
|
57
|
+
tenantId?: string,
|
|
58
|
+
options: MintMemoryCanaryOptions = {},
|
|
59
|
+
): MemoryCanaryEntry {
|
|
60
|
+
if (typeof id !== "string" || id.length === 0) {
|
|
61
|
+
throw new TypeError("mintMemoryCanary: 'id' must be a non-empty string");
|
|
62
|
+
}
|
|
63
|
+
if (typeof content !== "string") {
|
|
64
|
+
throw new TypeError("mintMemoryCanary: 'content' must be a string");
|
|
65
|
+
}
|
|
66
|
+
const tokenBytes = Math.max(
|
|
67
|
+
MIN_TOKEN_BYTES,
|
|
68
|
+
options.tokenBytes ?? DEFAULT_TOKEN_BYTES,
|
|
69
|
+
);
|
|
70
|
+
const canaryToken = randomBytes(tokenBytes).toString("hex");
|
|
71
|
+
const contentHash = computeHash(id, content, canaryToken, tenantId);
|
|
72
|
+
|
|
73
|
+
return {
|
|
74
|
+
id,
|
|
75
|
+
content,
|
|
76
|
+
contentHash,
|
|
77
|
+
canaryToken,
|
|
78
|
+
createdAt: new Date(),
|
|
79
|
+
tenantId,
|
|
80
|
+
};
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
/**
|
|
84
|
+
* Verify a previously minted canary against the content read back from
|
|
85
|
+
* storage. Returns `valid: true` only if both the content matches what
|
|
86
|
+
* was sealed and the tenant binding (if any) matches.
|
|
87
|
+
*
|
|
88
|
+
* Use case 1 — mutation detection:
|
|
89
|
+
* ```ts
|
|
90
|
+
* const entry = mintMemoryCanary("fact:42", "Sky is blue.");
|
|
91
|
+
* await db.write({ ...entry });
|
|
92
|
+
*
|
|
93
|
+
* // ... later ...
|
|
94
|
+
* const stored = await db.read("fact:42");
|
|
95
|
+
* const ver = verifyMemoryCanary(stored, stored.content);
|
|
96
|
+
* if (!ver.valid) {
|
|
97
|
+
* logger.security("Memory poisoning suspected", { id, reason: ver.reason });
|
|
98
|
+
* }
|
|
99
|
+
* ```
|
|
100
|
+
*
|
|
101
|
+
* Use case 2 — cross-tenant leak detection:
|
|
102
|
+
* ```ts
|
|
103
|
+
* // tenant A reads what should be a tenant-B-only entry
|
|
104
|
+
* const ver = verifyMemoryCanary(entry, entry.content, { tenantId: "tenant-A" });
|
|
105
|
+
* // ver.reason === "tenant_mismatch"
|
|
106
|
+
* ```
|
|
107
|
+
*/
|
|
108
|
+
export function verifyMemoryCanary(
|
|
109
|
+
entry: MemoryCanaryEntry,
|
|
110
|
+
observedContent: string,
|
|
111
|
+
options: { tenantId?: string } = {},
|
|
112
|
+
): MemoryCanaryVerification {
|
|
113
|
+
if (!entry || typeof entry !== "object") {
|
|
114
|
+
return { valid: false, reason: "canary_missing" };
|
|
115
|
+
}
|
|
116
|
+
if (
|
|
117
|
+
typeof entry.canaryToken !== "string" ||
|
|
118
|
+
entry.canaryToken.length < MIN_TOKEN_BYTES * 2
|
|
119
|
+
) {
|
|
120
|
+
return { valid: false, reason: "canary_missing" };
|
|
121
|
+
}
|
|
122
|
+
if (typeof observedContent !== "string") {
|
|
123
|
+
return { valid: false, reason: "content_mutated" };
|
|
124
|
+
}
|
|
125
|
+
// Tenant binding is fail-closed. If the entry was minted with a
|
|
126
|
+
// tenantId, the caller MUST supply the same tenantId. A caller that
|
|
127
|
+
// omits `options.tenantId` against a tenant-bound entry surfaces as
|
|
128
|
+
// a leak rather than a silent pass (Critic C2 from round 1 review).
|
|
129
|
+
if (entry.tenantId !== undefined) {
|
|
130
|
+
if (
|
|
131
|
+
options.tenantId === undefined ||
|
|
132
|
+
options.tenantId !== entry.tenantId
|
|
133
|
+
) {
|
|
134
|
+
return {
|
|
135
|
+
valid: false,
|
|
136
|
+
reason: "tenant_mismatch",
|
|
137
|
+
observed: observedContent,
|
|
138
|
+
};
|
|
139
|
+
}
|
|
140
|
+
} else if (
|
|
141
|
+
options.tenantId !== undefined &&
|
|
142
|
+
entry.tenantId !== options.tenantId
|
|
143
|
+
) {
|
|
144
|
+
// Caller passed a tenantId but the entry has none — also a mismatch.
|
|
145
|
+
return {
|
|
146
|
+
valid: false,
|
|
147
|
+
reason: "tenant_mismatch",
|
|
148
|
+
observed: observedContent,
|
|
149
|
+
};
|
|
150
|
+
}
|
|
151
|
+
|
|
152
|
+
const expectedHash = computeHash(
|
|
153
|
+
entry.id,
|
|
154
|
+
observedContent,
|
|
155
|
+
entry.canaryToken,
|
|
156
|
+
entry.tenantId,
|
|
157
|
+
);
|
|
158
|
+
if (!hashesEqual(expectedHash, entry.contentHash)) {
|
|
159
|
+
return {
|
|
160
|
+
valid: false,
|
|
161
|
+
reason:
|
|
162
|
+
observedContent === entry.content ? "hash_mismatch" : "content_mutated",
|
|
163
|
+
observed: observedContent,
|
|
164
|
+
};
|
|
165
|
+
}
|
|
166
|
+
return { valid: true };
|
|
167
|
+
}
|
|
168
|
+
|
|
169
|
+
/**
|
|
170
|
+
* Re-mint a canary after legitimate content edit. Returns a new
|
|
171
|
+
* sealed entry that supersedes the old one. The old `canaryToken`
|
|
172
|
+
* is rotated so a replay of the previous hash is also invalidated.
|
|
173
|
+
*/
|
|
174
|
+
export function rotateMemoryCanary(
|
|
175
|
+
prev: MemoryCanaryEntry,
|
|
176
|
+
newContent: string,
|
|
177
|
+
): MemoryCanaryEntry {
|
|
178
|
+
return mintMemoryCanary(prev.id, newContent, prev.tenantId);
|
|
179
|
+
}
|
|
180
|
+
|
|
181
|
+
/**
|
|
182
|
+
* Inject a *sentinel* memory entry — a decoy fact whose mutation would
|
|
183
|
+
* indicate the store was tampered with. Pair with `findSentinelMutations()`
|
|
184
|
+
* in a periodic sweep over the memory store.
|
|
185
|
+
*
|
|
186
|
+
* Returns a `MemoryCanaryEntry` with deterministic content that callers
|
|
187
|
+
* can recognise. The content includes the canary token so even content
|
|
188
|
+
* inspection (not just hash compare) catches mutation.
|
|
189
|
+
*/
|
|
190
|
+
export function buildSentinelEntry(
|
|
191
|
+
scope: string,
|
|
192
|
+
tenantId?: string,
|
|
193
|
+
): MemoryCanaryEntry {
|
|
194
|
+
// ID combines timestamp (for ordering) + random suffix (so enumeration
|
|
195
|
+
// by approximate-time guessing is infeasible — Critic M2 round 1).
|
|
196
|
+
const idSuffix = randomBytes(4).toString("hex");
|
|
197
|
+
const id = `ai-shield-sentinel:${scope}:${Date.now().toString(36)}-${idSuffix}`;
|
|
198
|
+
const nonce = randomBytes(8).toString("hex");
|
|
199
|
+
const content = `[AI-Shield sentinel — do not modify] scope=${scope} nonce=${nonce}`;
|
|
200
|
+
return mintMemoryCanary(id, content, tenantId);
|
|
201
|
+
}
|
|
202
|
+
|
|
203
|
+
/**
|
|
204
|
+
* Bulk verify a set of stored entries against their canaries. Returns
|
|
205
|
+
* the IDs that failed verification, with the reason.
|
|
206
|
+
*/
|
|
207
|
+
export function bulkVerify(
|
|
208
|
+
entries: Array<{
|
|
209
|
+
canary: MemoryCanaryEntry;
|
|
210
|
+
observedContent: string;
|
|
211
|
+
expectedTenantId?: string;
|
|
212
|
+
}>,
|
|
213
|
+
): Array<{ id: string; reason: NonNullable<MemoryCanaryVerification["reason"]> }> {
|
|
214
|
+
const failures: Array<{
|
|
215
|
+
id: string;
|
|
216
|
+
reason: NonNullable<MemoryCanaryVerification["reason"]>;
|
|
217
|
+
}> = [];
|
|
218
|
+
for (const e of entries) {
|
|
219
|
+
const v = verifyMemoryCanary(e.canary, e.observedContent, {
|
|
220
|
+
tenantId: e.expectedTenantId,
|
|
221
|
+
});
|
|
222
|
+
if (!v.valid && v.reason) {
|
|
223
|
+
failures.push({ id: e.canary.id, reason: v.reason });
|
|
224
|
+
}
|
|
225
|
+
}
|
|
226
|
+
return failures;
|
|
227
|
+
}
|
|
228
|
+
|
|
229
|
+
// --- Internal helpers ---
|
|
230
|
+
|
|
231
|
+
function computeHash(
|
|
232
|
+
id: string,
|
|
233
|
+
content: string,
|
|
234
|
+
token: string,
|
|
235
|
+
tenantId?: string,
|
|
236
|
+
): string {
|
|
237
|
+
return createHash("sha256")
|
|
238
|
+
.update(id)
|
|
239
|
+
.update("\0")
|
|
240
|
+
.update(content)
|
|
241
|
+
.update("\0")
|
|
242
|
+
.update(token)
|
|
243
|
+
.update("\0")
|
|
244
|
+
.update(tenantId ?? "")
|
|
245
|
+
.digest("hex");
|
|
246
|
+
}
|
|
247
|
+
|
|
248
|
+
/**
|
|
249
|
+
* Constant-time hex string compare. Falls back to a non-timing-safe
|
|
250
|
+
* direct compare only when lengths differ (which is itself the leak
|
|
251
|
+
* we want surfaced as a `false`, so this is fine).
|
|
252
|
+
*/
|
|
253
|
+
function hashesEqual(a: string, b: string): boolean {
|
|
254
|
+
if (typeof a !== "string" || typeof b !== "string") return false;
|
|
255
|
+
if (a.length !== b.length) return false;
|
|
256
|
+
// Safe — both buffers have identical length, both come from
|
|
257
|
+
// hex(SHA-256), no untrusted input.
|
|
258
|
+
return timingSafeEqual(Buffer.from(a, "hex"), Buffer.from(b, "hex"));
|
|
259
|
+
}
|