registry / circle-ir-ai / 2.40.1

circle-ir-ai@2.40.1

LLM-enhanced SAST analysis built on circle-ir

Static Scan Results

scanned 2h ago · by rust-scanner

Static analysis flagged 16 finding(s) at 72.0% confidence. This version is warn-only unless an AI or security-team review confirms malicious behavior.

Static reason

One or more suspicious static signals were detected.

Decision evidence

public snapshot

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsEvalFilesystemNetworkShell

Supply chain

HighEntropyStringsObfuscatedUrlStrings

ManifestNo manifest risk signals triggered.

scanned 169 file(s), 1.72 MB of source, external domains: apache.org, api.internal.com, api.npmjs.org, api.osv.dev, cwe.mitre.org, github.com, openvex.dev, raw.githubusercontent.com, registry.npmjs.org

Source & flagged code

7 flagged

dist/secret-scan/patterns.jsView file

Lines 209-267javascript

209 },

210 {

211 id: 'database-password',

212 name: 'Database Password',

213 description: 'Database password in configuration',

214 pattern: /(?:db[_-]?

Source & flagged code

7 flagged · loading source

dist/secret-scan/patterns.jsView file

229patternName = private_key_rsa severity = critical line = 229 matchedText = pattern:...-/g,

Critical

Critical Secret

Package contains a critical-looking secret pattern.

dist/secret-scan/patterns.jsView on unpkg · L229

229patternName = private_key_rsa severity = critical line = 229 matchedText = pattern:...-/g,

Critical

Secret Pattern

RSA private key in dist/secret-scan/patterns.js

dist/secret-scan/patterns.jsView on unpkg · L229

238patternName = private_key_openssh severity = critical line = 238 matchedText = pattern:...-/g,

Critical

Secret Pattern

OpenSSH private key in dist/secret-scan/patterns.js

dist/secret-scan/patterns.jsView on unpkg · L238

247patternName = private_key_ec severity = critical line = 247 matchedText = pattern:...-/g,

Critical

Secret Pattern

EC private key in dist/secret-scan/patterns.js

dist/secret-scan/patterns.jsView on unpkg · L247

dist/llm/language-context.jsView file

466**5. Code Evaluation** (CWE-94) - Operations that interpret data as code: L467: - Dynamic code execution, eval(), script engines L468: - Reflection with external class names

Low

Eval

Package source references a known benign dynamic code generation pattern.

dist/llm/language-context.jsView on unpkg · L466

dist/cache/tree-cache.jsView file

56try { L57: const require = createRequire(baseUrl); L58: return require(spec).version || 'unknown';

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/cache/tree-cache.jsView on unpkg · L56

dist/llm/discovery.jsView file

243// free skip. L244: const persistentCacheEnabled = process.env.LLM_DISCOVERY_CACHE !== '0'; L245: let dCache = null;

Low

Weak Crypto

Package source references weak cryptographic algorithms.

dist/llm/discovery.jsView on unpkg · L243

Findings

4 Critical4 Medium8 Low

CriticalCritical Secretdist/secret-scan/patterns.js

CriticalSecret Patterndist/secret-scan/patterns.js

MediumDynamic Requiredist/cache/tree-cache.js

MediumNetwork

MediumEnvironment Vars

MediumStructural Risk Force Deep Review

LowNon Install Lifecycle Scripts

LowScripts Present

LowEvaldist/llm/language-context.js

LowWeak Cryptodist/llm/discovery.js

LowFilesystem

LowObfuscated

LowHigh Entropy Strings

LowUrl Strings

dist/llm/language-context.jsView file

Lines 446-486javascript

446 - Process spawning, shell execution

447 - Script evaluation, macro execution

448 - ANY operation that interprets data as system commands

449 - KEY QUESTION: Can external data influence what command runs?

450 - SAFE: Fixed commands with data passed only as arguments (no shell interpretation)

451

452**3. Output Rendering** (CWE-79) - Operations that render content to users:

453 - HTML response generation, template rendering

454 - JavaScript output, DOM manipulation

455 - ANY operation that sends data to be rendered by clients

456 - KEY QUESTION: Can external data inject executable content?

457 - SAFE: Properly escaped output, JSON responses, auto-escaping templates

458

459**4. File System Operations** (CWE-22) - Operations that access file paths:

460 - File read/write, directory creation

461 - Path construction, file inclusion

462 - ANY operation that uses paths from external data

463 - KEY QUESTION: Can external data escape intended directories?

464 - SAFE: Validated paths, canonicalization with prefix checks

465

466**5. Code Evaluation** (CWE-94) - Operations that interpret data as code:

467 - Dynamic code execution, eval(), script engines

468 - Reflection with external class names

Low

Eval

Package source references a known benign dynamic code generation pattern.

dist/llm/language-context.jsView on unpkg · L466

469 - ANY operation that compiles/interprets data

470 - KEY QUESTION: Can external data become executable code?

471

472**6. Deserialization** (CWE-502) - Operations that reconstruct objects from data:

473 - Object deserialization, unmarshalling

474 - ANY operation that creates objects from serialized data

475 - KEY QUESTION: Can external data instantiate arbitrary classes?

476

477**7. Directory/Query Injection** (CWE-90, CWE-643) - Operations on hierarchical data:

478 - LDAP queries, XPath expressions

479 - ANY operation that queries tree/directory structures

480 - KEY QUESTION: Can external data modify query structure?

481

482**8. Network Requests** (CWE-918) - Operations that make outbound requests:

483 - HTTP clients, URL fetching, API calls

484 - ANY operation where URL/host is from external data

485 - KEY QUESTION: Can external data control request destination?

486

dist/cache/tree-cache.jsView file

Lines 36-76javascript

36 * **Env toggles:**

37 * LLM_TREE_CACHE=0 → disable persistent tree cache.

38 */

39import * as crypto from 'crypto';

40import * as fs from 'fs';

41import * as path from 'path';

42import { createRequire } from 'module';

43import { ClassificationCache } from './classification-cache.js';

44const TREE_CACHE_SUBDIR = path.join('.cognium', 'cache');

45const TREE_CACHE_TTL_MS = 7 * 24 * 60 * 60 * 1000; // 7 days — structural, stable

46/**

47 * Prompt-version sentinel. Tree-cache doesn't actually call an LLM

48 * (the input is pure SAST output), but reserving a prompt-version slot

49 * keeps the three caches' key compositions structurally identical and

50 * gives us a single dial to globally invalidate tree entries if the

51 * `PatternMatchResult` envelope ever changes shape.

52 */

53const PROMPT_VER = 'tree-v1';

54let _instance = null;

55function safeReadVersion(spec, baseUrl) {

56 try {

57 const require = createRequire(baseUrl);

58 return require(spec).version || 'unknown';

Medium

Dynamic Require

Package source references dynamic require/import behavior.

dist/cache/tree-cache.jsView on unpkg · L56

59 }

60 catch {

61 return 'unknown';

62 }

63}

64const _circleIrVersion = safeReadVersion('circle-ir/package.json', import.meta.url);

65const _circleIrAiVersion = (() => {

66 try {

67 const candidates = [

68 path.resolve(new URL(import.meta.url).pathname, '../../../package.json'),

69 path.resolve(new URL(import.meta.url).pathname, '../../package.json'),

70 ];

71 for (const p of candidates) {

72 if (fs.existsSync(p)) {

73 const pkg = JSON.parse(fs.readFileSync(p, 'utf8'));

74 if (pkg.name === 'circle-ir-ai' && pkg.version) {

75 return pkg.version;

76 }

dist/llm/discovery.jsView file

Lines 223-263javascript

223 }

224 return results;

225 }

226 /**

227 * Analyze a single method for vulnerabilities

228 */

229 async analyzeMethod(method, targetCWEs, options, irSinks = [], language = 'java') {

230 const methodId = `${method.className}.${method.methodName}`;

231 const confidenceThreshold = options.confidenceThreshold ?? 0.6;

232 // #163 B1 — canonicalize annotations + imports ONCE. The cache key

233 // and the LLM prompt MUST see the same view: same sort order, same

234 // top-20 cap. Otherwise two calls with permuted imports would

235 // produce the same key but different prompts (or vice versa),

236 // caching the wrong answer.

237 const canonicalAnnotations = [...new Set(method.annotations)].sort();

238 const canonicalImports = canonicalizeImportsForKey(method.imports);

239 // cognium-ai#134 — persistent discovery cache. Serves identical

240 // (methodCode, sortedCweSet, language, annotations, imports, model

241 // fingerprint) requests from disk so the 500× duplication pathology

242 // from the 2026-06-27 sweep (LLM-COST-SHARE-KEYCLOAK) becomes a

243 // free skip.

244 const persistentCacheEnabled = process.env.LLM_DISCOVERY_CACHE !== '0';

245 let dCache = null;

Low

Weak Crypto

Package source references weak cryptographic algorithms.

dist/llm/discovery.jsView on unpkg · L243

246 let dKey = null;

247 if (persistentCacheEnabled) {

248 dCache = getDiscoveryCache(getDefaultLLMConfig());

249 // #158 D6: className/methodName are intentionally NOT in the key

250 // (byte-identical bodies share answers; renames don't invalidate).

251 // #163 B1 (disc-v2): annotations + top-20 imports ARE in the key

252 // — they change the LLM's attack-surface reasoning and were

253 // silently colliding under disc-v1.

254 dKey = computeDiscoveryCacheKey({

255 methodCode: method.methodCode,

256 targetCWEs,

257 language,

258 annotations: canonicalAnnotations,

259 imports: canonicalImports,

260 });

261 const cached = await dCache.get(dKey.content, dKey.classifier);

262 if (cached) {

263 llmCallLogger.logDiscoveryCacheHit({