registry / Credential exfiltration / sagentic / 1.0.7

sagentic@1.0.7

Production-grade autonomous AI agent platform optimized for Google Cloud Run, Neon pgvector, and Gemini/Cloudflare

AI Security Review

scanned 3d ago · by lpm-firewall-ai

The CLI forwards local backend authentication secrets to a package-selected hosted endpoint during ordinary use, without a recipient-specific consent gate. This is concrete credential exfiltration behavior.

Static reason

High-risk behavior combination matched malicious policy.

Trigger

Invoke the packaged CLI in its default remote mode with AUTH_SECRET or API_SECRET set, then submit a prompt.

Impact

The endpoint receives a local backend credential that may enable unauthorized access to the user's backend.

Mechanism

The CLI selects the local secret as a bearer token and sends it in a POST request to the default hosted Sagentic endpoint.

Attack narrative

Running the packaged CLI starts its chat flow. In default remote mode, it selects AUTH_SECRET or API_SECRET from the local environment and sends the selected value as a bearer token to the fixed hosted endpoint. The transfer has no recipient-specific consent gate, exposing a credential defined for backend authentication.

AI rationale

Evidence

package.jsonbin/sagentic.jsdist/cli.jsdist/config/env.js

Network endpoints1

sagentic-ohyotaaora-el.a.run.app/api/v1/agent/chat

Decision evidence

public snapshot

AI called this Malicious at 94.0% confidence as Malware with low false-positive risk.

Evidence for block

The package exposes a CLI that starts the packaged runtime.
AUTH_SECRET is defined as a credential for HTTP backend authentication.
The CLI defaults to a fixed hosted Sagentic endpoint unless SAGENTIC_API_URL is set.
The remote request selects AUTH_SECRET or API_SECRET as a bearer token when higher-priority credentials are absent.
The CLI posts the selected token to the remote endpoint during a prompt.
Normal CLI prompts reach the remote request unless local mode is enabled.

Evidence against

The CLI supports an alternate endpoint and a local mode, but neither is required before sending the default credential.

Behavioral surface

Source

ChildProcessCryptoEnvironmentVarsEvalFilesystemNetworkShell

Supply chain

HighEntropyStringsUrlStrings

ManifestNo manifest risk signals triggered.

scanned 33 file(s), 143 KB of source, external domains: aistudio.google.com, api.cloudflare.com, cdn.tailwindcss.com, challenges.cloudflare.com, example.com, github.com, sagentic-ohyotaaora-el.a.run.app

Source & flagged code

5 flagged

dist/tools/data-calculator.jsView file

Lines 1-38javascript

1import { z } from "zod";

2const calculatorSchema = z.object({

3 expression: z

4 .string()

5 .describe('Arithmetic expression to compute, e.g. " (450 * 1.18) - 50 " or "Math.sqrt(144) + 25"'),

6 precision: z.coerce.number().int().min(0).max(10).default(4),

7});

8/**

9 * Evaluates mathematical expressions safely by parsing only numbers, math operators, and Math object methods.

10 * Strictly prohibits any access to process, window, global, require, or prototypes.

11 */

12function safeMathEval(expr) {

13 // Disallow any characters except numbers, whitespace, standard arithmetic operators, parentheses, commas, and Math.* functions

14 const sanitized = expr.replace(/\s+/g, "");

15 if (!/^([0-9\.\+\-\*\/\%\,]|Math\.(abs|acos|asin|atan|ceil|cos|exp|floor|log|max|min

16 throw new Error("Forbidden characters detected in math expression. Only standard arithmetic and Math methods are permitted.");

17 }

18 // Safe Function evaluation with no scope

19 const func = new Function(`return (${sanitized});`);

20 const result = func();

High

Eval

Package source references dynamic code evaluation.

dist/tools/data-calculator.jsView on unpkg · L18

21 if (typeof result !== "number" || isNaN(result)) {

22 throw new Error(`Expression did not evaluate to a valid number: ${expr}`);

23 }

24 return result;

25}

26export const dataCalculatorTool = {

27 name: "data_calculator",

28 description: "Safely calculates mathematical expressions, arithmetic, percentages, and scientific formulas with guaranteed precision.",

29 parameters: calculatorSchema,

30 execute: async ({ expression, precision }) => {

31 try {

32 const rawResult = safeMathEval(expression);

33 const rounded = Number(rawResult.toFixed(precision));

34 return {

35 expression,

36 result: rounded,

37 rawResult,

38 };

dist/security/auth.jsView file

Lines 1-29javascript

1import crypto from "crypto";

2import { env } from "../config/env.js";

3/**

4 * Constant-time string equality check to prevent side-channel timing attacks.

5 */

6export function safeEqual(

dist/cli.jsView file

•stage = ast_semantic_analysis; reason = ast_alias_growth_limit_exceeded; limitedFiles = 3

High

Semantic Analysis Limited

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

dist/cli.jsView on unpkg

Source & flagged code

5 flagged · loading source

dist/tools/data-calculator.jsView file

18// Safe Function evaluation with no scope L19: const func = new Function(`return (${sanitized});`); L20: const result = func();

High

Eval

Package source references dynamic code evaluation.

dist/tools/data-calculator.jsView on unpkg · L18

dist/security/auth.jsView file

9} L10: const bufA = Buffer.from(a); L11: const bufB = Buffer.from(b); ... L30: !safeEqual(originSecretHeader, env.ORIGIN_SECRET)) { L31: reply.status(403).send({ L32: statusCode: 403, ... L62: : false; L63: const matchesApiSecret = process.env.API_SECRET L64: ? safeEqual(token, process.env.API_SECRET) ... L121: } L122: const res = await fetch("https://challenges.cloudflare.com/turnstile/v0/siteverify", { L123: method: "POST",

Critical

Credential Exfiltration

Source appears to send environment or credential material to an external endpoint.

dist/security/auth.jsView on unpkg · L9

9Trigger-reachable chain: scripts.start -> dist/server.js -> dist/routes/agent.js -> dist/security/auth.js L9: } L10: const bufA = Buffer.from(a); L11: const bufB = Buffer.from(b); ... L30: !safeEqual(originSecretHeader, env.ORIGIN_SECRET)) { L31: reply.status(403).send({ L32: statusCode: 403, ... L62: : false; L63: const matchesApiSecret = process.env.API_SECRET L64: ? safeEqual(token, process.env.API_SECRET) ... L121: } L122: const res = await fetch("https://challenges.cloudflare.com/turnstile/v0/siteverify", { L123: method: "POST",

Findings

2 Critical3 High3 Medium4 Low

CriticalCredential Exfiltrationdist/security/auth.js

CriticalTrigger Reachable Dangerous Capabilitydist/security/auth.js

HighEvaldist/tools/data-calculator.js

HighTrigger Reachable Credential Exfiltrationdist/security/auth.js

HighSemantic Analysis Limiteddist/cli.js

MediumNetwork

MediumEnvironment Vars

MediumStructural Risk Force Deep Review

LowScripts Present

LowFilesystem

LowHigh Entropy Strings

LowUrl Strings

Affected versions and remediation

This report applies to sagentic@1.0.7.

See all versions of sagentic for other recorded verdicts.

Avoid installing sagentic@1.0.7. Remove it from direct dependencies and check your lockfile for transitive copies.
Choose an independently verified alternative or release. This report does not establish that other versions are safe.
If this version ran, investigate the affected machine and build environment. Rotate credentials it could access and rebuild from a trusted environment.

Evidence last updated: 2026-10-20 05:55:44.570Z (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

Related packages

Other credential exfiltration packages

Behavioral surface

Source & flagged code

Related packages

Other credential exfiltration packages