registry / AI-agent capability abuse / @livx.cc/agentx / 0.99.33

@livx.cc/agentx@0.99.33

Edge-native AI agent runtime — drives a virtual filesystem via any LLM (ai.libx.js). Same bytes run in node, browser, or edge.

AI Security Review

scanned 2mo ago · by lpm-firewall-ai

The package is a dangerous dual-use agent runtime with real shell and persistence features, warranting a warning. Direct inspection found no malicious lifecycle behavior or concrete exfiltration chain.

Static reason

No blocking static signals were detected.; previous stored version diff introduced dangerous source

Trigger

A user runs the agentx CLI and supplies a task or configuration.

Impact

A prompt, hostile project configuration, or model error can cause commands and file changes with the invoking user's privileges.

Mechanism

User-invoked AI agent with shell execution, writable project state, and optional scheduled persistence.

AI rationale

Evidence

package.jsondist/cli.jsdist/tools.shell.jscurrent working directory./.agent/config.{ts,js,mjs,json}./.agent/sessions/~/.agent/sessions/

Network endpoints4

api.anthropic.com/v1/messagesapi.tavily.com/searchapi.firecrawl.dev/v1/searchhtml.duckduckgo.com/html/

Decision evidence

public snapshot

AI called this Suspicious at 93.0% confidence as Dangerous Capability with low false-positive risk.

Evidence for warning

The user-invoked CLI creates a real-disk AI agent and exposes arbitrary shell command execution.
The CLI can schedule prompts through the operating system, resuming headlessly after exit.
When invoked, it dynamically imports project or user agent configuration JavaScript before the trust prompt.

Evidence against

Registry consumers have no preinstall, install, or postinstall hook; prepare is a VCS-install hook and prepublishOnly is publish-time.
Library import does not invoke the CLI main function.
No package self-dependency, fixed credential-redirection gateway, foreign-package overwrite, or unconsented install-time mutation was found.

Behavioral surface

Source

ChildProcessCryptoDynamicRequireEnvironmentVarsEvalFilesystemNetworkShellWebSocket

Supply chain

HighEntropyStringsPossibleObfuscation

Source & flagged code

8 flagged

dist/index.jsView file

Lines 4957-4997javascript

4957 ["setImmediate", /\bsetImmediate\b/],

4958 ["Atomics", /\bAtomics\b/],

4959 ["SharedArrayBuffer", /\bSharedArrayBuffer\b/],

4960 ["WebAssembly", /\bWebAssembly\b/],

4961 ["Reflect", /\bReflect\b/],

4962 ["Proxy", /\bProxy\b/],

4963 ["Buffer", /\bBuffer\b/],

4964 ["while(true)", /while\s*$\s*true\s*$/],

4965 ["for(;;)", /for\s*$\s*;\s*;\s*$/],

4966 // escape sequences can smuggle a denied identifier past the substring check

4967 // (e.g. `process` parses as `process`), so forbid them outright.

4968 ["\\u escape", /\\u/],

4969 ["\\x escape", /\\x/]

4970 ];

4971 const violations = DENY.filter(([, re]) => re.test(code)).map(([name]) => name);

4972 return { ok: violations.length === 0, violations };

4973}

4974function compileSynthesizedTool(spec) {

4975 const v = validateToolCode(spec.code);

4976 if (!v.ok) throw new Error(`unsafe synthesized tool '${spec.name}': blocked tokens [${v.violations.join(", ")}]`);

4977 if (!/^[A-Za-z_]\w{0,39}$/.test(spec.name)) throw new Error(`invalid tool name '${spec.name}'`);

4978 const fn = new Function("args", "ctx", `"use strict"; return (async (args, ctx) => { ${spec.code} })(args, ctx);`);

4979 return {

Low

Eval

Package source references a known benign dynamic code generation pattern.

dist/index.jsView on unpkg · L4977

4980 name: spec.name,

4981 description: spec.description,

4982 parameters: spec.parameters ?? { type: "object", properties: {} },

4983 async run(args, ctx) {

4984 const r = await fn(args, ctx);

4985 return typeof r === "string" ? r : JSON.stringify(r ?? "");

4986 }

4987 };

4988}

4989

4990// src/FakeAIClient.ts

4991var FakeAIClient = class {

4992 seen = [];

4993 script;

4994 constructor(script) {

4995 this.script = [...script];

4996 }

4997 async chat(options) {

dist/cli.jsView file

•Manifest-reachable source captures an API credential, sends it to a fixed unofficial gateway, and persists that redirection in agent or shell configuration. dist/cli.js: await ctx.fs.writeFile(path, body); await ctx.fs.writeFile(path, current); await ctx.fs.writeFile(p, content); headers: { authorization: `Bearer ${opts.key}`, "content-type": "application/json" }, async writeFile(path, content) { for (const [p, c] of this.st.writes) await this.base.writeFile(p, c); await ctx.fs.writeFile(path, next); return join2(repoRoot, ".claude", "worktrees");

High

Credential Redirect Persistence

Manifest-reachable source sends a prompted API credential to a fixed unofficial gateway and persists the redirection.

dist/cli.jsView on unpkg

•Manifest-reachable source resolves another installed package, overwrites its runtime code, and injects package-defined remote behavior. dist/cli.js: DEFAULT_SKIP_DIRS = ["node_modules", ".git", "dist", ".tmp"]; await ctx.fs.writeFile(path, body); await ctx.fs.writeFile(path, current); await ctx.fs.writeFile(p, content); res = await doFetch(current, { signal: ctl.signal, redirect: "manual", headers: { "user-agent": "agentx (+https://github.com/Livshitz/agentx)" } }); const res = await opts.fetch("https://api.anthropic.com/v1/messages", { const tavilyEndpoint = options.endpoint ?? "https://api.tavily.com/search"; const firecrawlEndpoint = options.firecrawlEndpoint ?? "https://api.firecrawl.dev/v1/search";

High

Entrypoint Foreign Package Code Overwrite

Manifest-reachable source overwrites another installed package with package-defined remote behavior.

dist/cli.jsView on unpkg

35Trigger-reachable persistence chain: manifest.bin -> dist/cli.js L35: if (content == null || typeof content === "string") return false; L36: return content.some((p) => p.type === "image_url" && (p.image_url?.url ?? "").startsWith("data:")); L37: } ... L156: try { L157: return parseImageRef(JSON.parse(content).imageRef); L158: } catch (e) { ... L295: CONFIG_FILE_RE = /(^|\/)\.(agent|claude)\/(settings(\.[\w-]+)?\.json|config\.(json|js|mjs|cjs|ts))$/i; L296: SECRET_PAIR = /((?:^|[\s,{[])(?:export\s+)?["']?[\w.\-]*(?:KEY|TOKEN|SECRET|PASSWORD|PASSWD|CREDENTIAL|PRIVATE_KEY|ACCESS_?KEY|AUTH(?:_?TOKEN)?|BEARER)[\w.\-]*["']?\s*[:=]\s*)(["']... L297: SECRET_TOKEN = /\b(sk-ant-[\w-]{12,}|sk-[A-Za-z0-9]{20,}|ghp_[A-Za-z0-9]{20,}|gho_[A-Za-z0-9]{20,}|github_pat_[\w]{20,}|xox[baprs]-[\w-]{10,}|AKIA[0-

package.jsonView file

Lines 43-83json

43 "runtime",

44 "tools",

45 "mcp",

46 "claude",

47 "sandbox"

48 ],

49 "engines": {

dist/tools.shell.jsView file

•matchType = previous_version_dangerous_delta matchedPackage = @livx.cc/agentx@0.99.31 matchedIdentity = npm:QGxpdnguY2MvYWdlbnR4:0.99.31 similarity = 0.400 summary = stored previous version shares package body but lacks this dangerous source file

High

Previous Version Dangerous Delta

This package version adds a dangerous source file absent from the previous stored version; route for source-aware review.

dist/tools.shell.jsView on unpkg

Source & flagged code

8 flagged · loading source

dist/index.jsView file

4977if (!/^[A-Za-z_]\w{0,39}$/.test(spec.name)) throw new Error(`invalid tool name '${spec.name}'`); L4978: const fn = new Function("args", "ctx", `"use strict"; return (async (args, ctx) => { ${spec.code} })(args, ctx);`); L4979: return {

Low

Eval

Package source references a known benign dynamic code generation pattern.

dist/index.jsView on unpkg · L4977

dist/cli.jsView file

High

Credential Redirect Persistence

Manifest-reachable source sends a prompted API credential to a fixed unofficial gateway and persists the redirection.

dist/cli.jsView on unpkg

Findings

4 High7 Medium6 Low

HighCredential Redirect Persistencedist/cli.js

HighEntrypoint Foreign Package Code Overwritedist/cli.js

HighTrigger Reachable Persistencedist/cli.js

HighPrevious Version Dangerous Deltadist/tools.shell.js

MediumDynamic Requiredist/cli.js

MediumNetwork

MediumEnvironment Vars

MediumInstall Persistencedist/cli.js

MediumStructural Risk Force Deep Review

MediumRemote Tarball Dependencypackage.json

MediumWildcard Dependency

LowNon Install Lifecycle Scripts

LowScripts Present

LowEvaldist/index.js

Affected versions and remediation

This report applies to @livx.cc/agentx@0.99.33.

See all versions of @livx.cc/agentx for other recorded verdicts.

Review the evidence and your use of @livx.cc/agentx@0.99.33 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: 2026-08-28 21:36:20.913Z (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