registry / AI-agent capability abuse / @nordsym/apiclaw / 2.9.7

@nordsym/apiclaw@2.9.7

Your agent calls real APIs. You sign in once. Authenticated execution and discovery layer for AI agents, reachable via SKILL.md, local MCP, CLI, HTTP gateway, or Remote MCP. Discover 26,619 API definitions, including 689 exact-name source-verified entries

AI Security Review

scanned 2mo ago · by lpm-firewall-ai

The source establishes automatic identifiable telemetry plus explicit AI-agent configuration and agent-directed command behavior. It does not establish the unconsented install-time foreign control-surface mutation or payload execution required for a publish block.

Static reason

High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review; previous stored version diff introduced dangerous source

Trigger

npm installation triggers telemetry; running setup or mcp-install changes MCP configuration.

Impact

Tracks an installing machine and can add a package-controlled MCP capability to an AI client.

Mechanism

Install telemetry and explicit MCP self-registration.

AI rationale

Evidence

dist/postinstall.jsdist/funnel-client.jsdist/session.jsdist/cli/commands/mcp-install.jsdist/index.js~/.apiclaw/funnel-markers.json~/.claude.json~/.codex/config.toml

Network endpoints1

adventurous-avocet-799.convex.cloud/api/mutation

Decision evidence

public snapshot

AI called this Suspicious at 88.0% confidence as Dangerous Capability with low false-positive risk.

Evidence for warning

The postinstall hook sends an install event containing a hostname-and-user-name fingerprint without an affirmative prompt.
The explicit MCP installer writes agent configuration that launches npx with this package name, allowing later resolution and lifecycle execution.
The MCP tool description gives imperative instructions to an AI agent to drive browser authentication and repeatedly run package commands.

Evidence against

Postinstall does not write AI-agent configuration; it only records telemetry and prints setup instructions.
Agent configuration changes are exposed through explicit setup and mcp-install CLI commands, including dry-run options.
No source inspected shows credential harvesting, shell payload download, or destructive actions.

Behavioral surface

Source

ChildProcessCryptoEnvironmentVarsFilesystemNetworkShell

Supply chain

HighEntropyStringsMinified

Source & flagged code

14 flagged

package.jsonView file

Lines 1-41json

2 "name": "@nordsym/apiclaw",

3 "version": "2.9.7",

4 "description": "Your agent calls real APIs. You sign in once. Authenticated execution and discovery layer for AI agents, reachable via SKILL.md, local MCP, CLI, HTTP gateway, or Remote MCP. Discover 26,619 API definitions, including 689 exact-name source-verified entries. Source verification is not execution. APIClaw inventories 22 built-in providers, all customer-executable today, plus 1,003 workspace-authenticated public/no-key HTTPS origins (1,025 callable now). Anonymous keyless proxy stays disabled. Agent-native auth: `apiclaw auth login` opens your browser, signs you in, and writes ~/. ...

5 "type": "module",

6 "main": "dist/index.js",

7 "homepage": "https://apiclaw.cloud",

8 "repository": {

9 "type": "git",

10 "url": "https://github.com/nordsym/apiclaw"

11 },

12 "bugs": {

13 "url": "https://github.com/nordsym/apiclaw/issues"

14 },

15 "files": [

16 "dist/"

17 ],

18 "scripts": {

19 "build": "node -e \"require('node:fs').rmSync('dist', { recursive: true, force: true })\" && ./node_modules/.bin/tsc && chmod +x dist/bin.js",

20 "prepublishOnly": "npm run build",

21 "postinstall": "node dist/postinstall.js || true",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L21

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg · L21

22 "dev": "tsx watch src/index.ts",

23 "start": "node dist/index.js",

24 "test": "tsx src/test.ts && tsx src/first-run.test.ts && tsx src/first-execute-rails.test.ts && tsx src/first-call.test.ts && tsx src/execute-auth.test.ts && tsx src/auth-config.test.ts && tsx src/login-wait.test.ts && tsx src/cli-auth-redeem.test.ts && tsx src/pending-login-start.test.ts && tsx src/skill.test.ts && tsx src/funnel.test.ts && tsx src/product-truth.test.ts && tsx src/workspace-public-apis.test.ts && tsx src/apilayer-customer-rails.test.ts && tsx src/version-trap.test.ts && tsx src/mcp-scope-policy.test.ts && tsx src/trust-floor.test.ts && tsx src/provider-boundaries.test.ts ...

25 "audit:managed-providers": "node scripts/audit-managed-providers.mjs",

26 "test:e2e": "tsx src/test.ts",

27 "webhook": "tsx src/webhook.ts",

28 "convex:dev": "npx convex dev",

29 "convex:deploy": "find convex -maxdepth 1 \\( -name '*.js' -o -name '*.js.map' -o -name '*.d.ts' -o -name

30 },

31 "keywords": [

32 "mcp",

33 "model-context-protocol",

34 "ai-agent",

35 "ai-agents",

36 "claude",

37 "gpt",

38 "llm",

39 "api",

40 "api-gateway",

41 "api-discovery",

Long lines were clipped for display.

dist/skill.test.jsView file

Lines 1-28javascript

1#!/usr/bin/env npx tsx

2/**

3 * Agent front door. An agent given

4 * `set up https://apiclaw.cloud/SKILL.md` must be able to install, auth,

5 * and land one POST /v1/execute without guessing. llms.txt and /api/catalog

6 * already exist; SKILL.md is the operational door, not a second canon.

7 */

8import assert from "node:assert/strict";

9import

dist/pending-login-start.jsView file

Lines 1-24javascript

1/**

2 * Mint (or reuse) a pending Clerk CLI login without a TTY.

3 *

4 * MCP / agent hosts have no stdin/stdout TTY, so they must not wait for

5 * `canLaunchInteractiveAuth` before creating `https://apiclaw.cloud/auth/cli?authId=…`.

6 * Opening the browser is best-effort. whoami redeems Authorize; localhost

7 * connection refused is OK.

8 */

9import { createHash, randomBytes } from "node:crypto"

dist/postinstall.jsView file

Lines 1-21javascript

1#!/usr/bin/env node

2/**

3 * APIClaw Postinstall Hook

4 * Prints a welcome message with links and fires ONE anonymous funnel

5 * event (install) so we can measure top-of-funnel truthfully.

6 *

7 * Disable telemetry: APICLAW_TELEMETRY=false

8 */

9import { emitFunnelEvent, hasLocalMarker, setLocalMarker } from './funnel-client.js';

dist/cli/commands/mcp-install.jsView file

Lines 1-28javascript

1/**

2 * MCP Install Command

3 * Simple, focused command to install APIClaw into MCP config files

4 * Supports Claude Desktop and Claude Code

5 */

6import { existsSync, readFileSync, writeFileSync, mkdirSync } from 'fs';

7import { dirname, join } from 'path';

dist/cli/commands/direct.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; 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/commands/direct.js#virtual:normalized:round1View on unpkg

dist/acp-server.test.jsView file

•matchType = normalized_sha256 matchedPackage = @nordsym/apiclaw@2.9.6 matchedPath = dist/acp-server.test.js matchedIdentity = npm:QG5vcmRzeW0vYXBpY2xhdw:2.9.6 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High

Known Malware Source Similarity

Source file is highly similar to a previously finalized malicious package; route for source-aware review.

dist/acp-server.test.jsView on unpkg

dist/cli/commands/doctor.jsView file

•matchType = normalized_sha256 matchedPackage = @nordsym/apiclaw@2.9.6 matchedPath = dist/cli/commands/doctor.js matchedIdentity = npm:QG5vcmRzeW0vYXBpY2xhdw:2.9.6 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High

Known Malware Source Similarity

Source file is highly similar to a previously finalized malicious package; route for source-aware review.

dist/cli/commands/doctor.jsView on unpkg

dist/cli/commands/direct.jsView file

•matchType = normalized_sha256 matchedPackage = @nordsym/apiclaw@2.9.6 matchedPath = dist/cli/commands/direct.js matchedIdentity = npm:QG5vcmRzeW0vYXBpY2xhdw:2.9.6 similarity = 1.000 summary = normalized source hash matched finalized malicious source

High

Known Malware Source Similarity

Source file is highly similar to a previously finalized malicious package; route for source-aware review.

dist/cli/commands/direct.jsView on unpkg

Source & flagged code

14 flagged · loading source

package.jsonView file

•scripts.postinstall = node dist/postinstall.js || true

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.postinstall = node dist/postinstall.js || true

Medium

Ambiguous Install Lifecycle Script

Install-time lifecycle script is not statically allowlisted and needs review.

package.jsonView on unpkg

dist/skill.test.jsView file

8import assert from "node:assert/strict"; L9: import { spawnSync } from "node:child_process"; L10: import { existsSync, mkdtempSync, readFileSync, rmSync } from "node:fs";

High

Child Process

Package source references child process execution.

dist/skill.test.jsView on unpkg · L8

Findings

2 Critical12 High4 Medium5 Low

CriticalAi Agent Control Hijackdist/postinstall.js

CriticalPrevious Version Dangerous Deltadist/cli/commands/mcp-install.js

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processdist/skill.test.js

HighShell

HighSame File Env Network Executiondist/pending-login-start.js

HighSandbox Evasion Gated Capabilitydist/pending-login-start.js

HighRuntime Package Installdist/cli/commands/mcp-install.js

HighSemantic Analysis Limiteddist/cli/commands/direct.js#virtual:normalized:round1

HighKnown Malware Source Similaritydist/cli/commands/mcp-install.js

HighKnown Malware Source Similaritydist/pending-login-start.js

Affected versions and remediation

This report applies to @nordsym/apiclaw@2.9.7.

See all versions of @nordsym/apiclaw for other recorded verdicts.

Review the evidence and your use of @nordsym/apiclaw@2.9.7 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: 2026-08-31 21:01:57.219Z (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