registry / Credential exfiltration / @ngockhoale/ukit / 3.4.23

@ngockhoale/ukit@3.4.23

Install/update an index-first AI workspace for Claude Code, OpenAI Codex and omp (Oh My Pi).

AI Security Review

scanned 15d ago · by lpm-firewall-ai

Source inspection confirms that decision requests can forward an existing Claude gateway credential to an unrestricted configured endpoint. The feature is stage-gated, so this supports a warning for conditional credential exposure rather than a finding of covert theft or malware.

Static reason

No blocking static signals were detected.

Trigger

A task-routing call with the decision-plane stage enabled and a configured decision endpoint.

Impact

A configured endpoint outside the credential's intended service could receive the credential.

Mechanism

The client resolves the endpoint independently from the Claude gateway URL, falls back to the Claude gateway credential when no decision-specific key exists, and sends that credential in the request headers. The endpoint is not restricted by an allowlist in the inspected code.

AI rationale

Evidence

src/core/gatewayProbe.jssrc/decision/client.jssrc/index/taskRouting.js

Network endpoints1

openai.unicjsc.com

Decision evidence

public snapshot

AI called this Suspicious at 92.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The decision endpoint can come directly from the operator's UKIT_DECISION_BASE_URL without a destination allowlist.
When no decision-specific key is set, the client falls back to the existing Claude gateway credential, including ANTHROPIC_AUTH_TOKEN.
The client places that credential in an Authorization or API key header and sends it in a POST request to the selected endpoint.
Task routing activates decision requests when the configured decision-plane stage is not off.

Evidence against

The decision-plane stage defaults to off when absent or malformed, and routing can retain deterministic behavior when it is disabled.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemShell

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 259 file(s), 2.97 MB of source, external domains: openai.unicjsc.com

Source & flagged code

6 flagged

scripts/bench/runGold.mjsView file

Lines 101-141javascript

101 failures.push(`budget: used ${packet.budget.used} > maxTokens ${maxTokens}`);

102 }

103

104 return {

105 id: testCase.id ?? '(unnamed)',

106 mode: packet.task_type,

107 expectedMode,

108 anchors: [...anchorPaths],

109 expectedAnchors,

110 anchorHits,

111 anchorMisses,

112 budgetUsed: packet.budget?.used ?? 0,

113 maxTokens,

114 routedMode: routed.mode,

115 pass: failures.length === 0,

116 failures,

117 };

118}

119

120export async function runGold({ projectRoot, corpusPath } = {}) {

121 const rootDir = projectRoot ?? REPO_ROOT;

122 const { compileContext } = await import(path.join(REPO_ROOT, 'src/core/codeintel/compiler.js'));

123 const { routeTask } = await import(path.join(REPO_ROOT, 'src/core/codeintel/router.js'));

Medium

Dynamic Require

Package source references dynamic require/import behavior.

scripts/bench/runGold.mjsView on unpkg · L121

124 const { buildCodeIndex } = await import(path.join(REPO_ROOT, 'src/index/buildIndex.js'));

125 const { INDEX_ARTIFACTS, getArtifactPath } = await import(path.join(REPO_ROOT, 'src/index/paths.js'));

126

127 // The bench needs an index to score anchors; build once when missing.

128 try {

129 await fs.stat(getArtifactPath(rootDir, INDEX_ARTIFACTS.files));

130 } catch {

131 console.log('[bench] index missing — building code index first');

132 await buildCodeIndex({ rootDir });

133 }

134

135 const cases = await loadCorpus(corpusPath ?? DEFAULT_CORPUS);

136 const results = [];

137 for (const testCase of cases) {

138 results.push(await runGoldCase(rootDir, testCase, { compileContext, routeTask }));

139 }

140 return { results, pass: results.every((r) => r.pass) };

141}

scripts/release/verify-release.mjsView file

Lines 1-21javascript

1import { spawn } from 'node:child_process';

2import { readFileSync, writeFileSync, unlinkSync, existsSync } from 'node:fs';

3import crypto from 'node:crypto';

4import os from 'node:os'

template_project/.claude/hooks/task-watchdog.shView file

•path = template_project/.claude/hooks/task-watchdog.sh kind = payload_in_excluded_dir sizeBytes = 11622 magicHex = [redacted]

High

Payload In Excluded Dir

Package hides binary, compressed, or executable-looking payloads in test/fixture/hidden paths.

template_project/.claude/hooks/task-watchdog.shView on unpkg

•path = template_project/.claude/hooks/task-watchdog.sh kind = build_helper sizeBytes = 11622 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

template_project/.claude/hooks/task-watchdog.shView on unpkg

scripts/audit/decision-coverage.mjs#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 11

High

Semantic Analysis Limited

A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.

scripts/audit/decision-coverage.mjs#virtual:normalized:round1View on unpkg

src/core/gatewayProbe.jsView file

Lines 41-81javascript

42function readUserDecisionBlock(homeDir) {

43 if (!homeDir) return {};

44 const g = safeReadJson(path.join(homeDir,

Source & flagged code

6 flagged · loading source

scripts/bench/runGold.mjsView file

121const rootDir = projectRoot ?? REPO_ROOT; L122: const { compileContext } = await import(path.join(REPO_ROOT, 'src/core/codeintel/compiler.js')); L123: const { routeTask } = await import(path.join(REPO_ROOT, 'src/core/codeintel/router.js'));

Medium

Dynamic Require

Package source references dynamic require/import behavior.

scripts/bench/runGold.mjsView on unpkg · L121

scripts/release/verify-release.mjsView file

1import { spawn } from 'node:child_process'; L2: import { readFileSync, writeFileSync, unlinkSync, existsSync } from 'node:fs'; ... L6: L7: const rootDir = process.cwd(); L8: const npmCacheDir = path.join(os.tmpdir(), 'ukit-npm-cache'); ... L14: // UKIT_VERIFY_STEP_TIMEOUT_MS overrides the step budget for tests. L15: const STEP_TIMEOUT_MS = Number(process.env.UKIT_VERIFY_STEP_TIMEOUT_MS) || 10 * 60 * 1000; // 10 min L16: const REGISTRY_TIMEOUT_MS = Number(process.env.UKIT_VERIFY_REGISTRY_TIMEOUT_MS) || 60 * 1000; // 1 min ... L18: // Opt-in post-publish registry-parity check (Release Policy invariant: npm and git always hold L19: // the same latest version). Only meaningful AFTER `npm publish` — before publish the registry is L20: // behind by definition — so it is gated behind --post-

Low

Weak Crypto

Package source references weak cryptographic algorithms.

Findings

2 High5 Medium5 Low

HighPayload In Excluded Dirtemplate_project/.claude/hooks/task-watchdog.sh

HighSemantic Analysis Limitedscripts/audit/decision-coverage.mjs#virtual:normalized:round1

MediumDynamic Requirescripts/bench/runGold.mjs

MediumEnvironment Vars

MediumShips Build Helpertemplate_project/.claude/hooks/task-watchdog.sh

MediumStructural Risk Force Deep Review

MediumAi Review Evidencesrc/core/gatewayProbe.js

LowScripts Present

LowWeak Cryptoscripts/release/verify-release.mjs

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

Affected versions and remediation

This report applies to @ngockhoale/ukit@3.4.23.

See all versions of @ngockhoale/ukit for other recorded verdicts.

Review the evidence and your use of @ngockhoale/ukit@3.4.23 before allowing it. Restrict the permissions described in this report.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.

Evidence last updated: 2026-10-08 01:02:59.009Z (UTC).

Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code