registry / AI-agent capability abuse / orquesta-agent / 0.2.256

orquesta-agent@0.2.256

Local agent for Orquesta - connects your VM to the Orquesta dashboard

AI Security Review

scanned 27d ago · by lpm-firewall-ai

Install hooks only print messages, but explicit init and token connect still write foreign Claude and Cursor control files and can drop server-supplied skills and OAuth credentials onto disk. That is a guarded, user-invoked agent capability, so the package should be warned rather than blocked.

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

A user runs orquesta-agent init with a token, or starts the agent with a token so it connects and syncs project skills.

Impact

Later Claude and Cursor sessions run this package's hook handler and can follow dashboard-supplied skill and CLAUDE.md text, and can use server-supplied Claude credentials stored in the user profile.

Mechanism

init merges command hooks into .claude/settings.json and .cursor/hooks.json. The connect path downloads skills and credentials from the Orquesta API and writes them into the working tree and, when supplied, ~/.claude/.credentials.json.

AI rationale

Evidence

dist/init.jsdist/index.jsdist/supabase.jsdist/executor.jspackage.json.claude/settings.json.cursor/hooks.json.claude/skills.claude/CLAUDE.mdCLAUDE.md~/.claude/.credentials.json.orquesta.json

Network endpoints2

getorquesta.comapuchat.com

Decision evidence

public snapshot

AI called this Suspicious at 78.0% confidence as Dangerous Capability with medium false-positive risk.

Evidence for warning

The init CLI command writes Claude Code hook commands into the project .claude/settings.json so later prompts and tool use invoke orquesta-agent.
The same init path, defaulting to both clients, also merges hook commands into .cursor/hooks.json.
After a token connect, fetchSkills writes server-supplied skill files under the project .claude/skills directory and can write or append CLAUDE.md.
When the server returns Claude OAuth material, configureAuth writes it to ~/.claude/.credentials.json in the user home directory.
preinstall and postinstall only check the Node version and print install banners; they do not change agent settings.

Evidence against

Hook setup requires the explicit init command and a --token flag; it is not an npm install hook.
The package does not depend on its own name, so there is no nested lifecycle self-install chain.
Hook payloads are passed through redactDeep before they are sent to the configured API.
Credential and skill content is fetched from the Orquesta API with the operator token and then applied locally, rather than scraped from unrelated services.

Source & flagged code

22 flagged

package.jsonView file

Lines 1-38json

2 "name": "orquesta-agent",

3 "version": "0.2.256",

4 "description": "Local agent for Orquesta - connects your VM to the Orquesta dashboard",

5 "type": "module",

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

7 "bin": {

8 "orquesta-agent": "./dist/index.js"

9 },

10 "scripts": {

11 "build": "tsc && cp -r src/ui/public dist/ui/",

12 "build:bundle": "npm run build && npx esbuild dist/index.js --bundle --platform=node --target=node20 --format=esm --outfile=bin/orquesta-agent.mjs --external:@homebridge/node-pty-prebuilt-multiarch --banner:js=\"import{createRequire}from'module';const require=createRequire(import.meta.url);\" --minify",

13 "build:binary": "npm run build && bash scripts/build-binaries.sh",

14 "dev": "tsc --watch",

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

16 "prepublishOnly": "npm run build:binary",

17 "preinstall": "node scripts/preinstall.js",

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg · L17

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg · L17

18 "postinstall": "node scripts/postinstall.js"

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg · L18

19 },

20 "keywords": [

21 "orquesta",

22 "agent",

23 "claude",

24 "ai",

25 "automation"

26 ],

27 "author": "Orquesta",

28 "license": "MIT",

29 "dependencies": {

30 "@anthropic-ai/sdk": "^0.95.1",

31 "@homebridge/node-pty-prebuilt-multiarch": "^0.13.1",

32 "@nut-tree-fork/nut-js": "4.2.6",

33 "@types/cors": "^2.8.19",

34 "@types/express": "^5.0.6",

35 "chalk": "^5.3.0",

36 "commander": "^11.1.0",

37 "cors": "^2.8.6",

38 "express": "^5.2.1",

dist/ui/onboarding.jsView file

Lines 1-33javascript

1// First-run onboarding: the guided welcome experience that opens on the user's

2// machine the first time the agent connects after install. It explains what the

3// agent is, the execution modes (LLM engine / permissions / sandbox) and the

4// localhost:8080 manager UI — and lets the user apply those settings live.

5//

6// Design: the agent process itself serves the welcome page (embedded UI server)

7// so the "Configure now" step can apply changes to the RUNNING agent with no

8// restart. We open the browser only when a local display is present; on headless

9// servers (VMs, customer boxes over SSH) we just print a prominent URL banner.

dist/index.jsView file

Lines 1-35javascript

1#!/usr/bin/env node

2// Node.js version check — must run before any imports that use fetch/etc.

3const [major] = process.versions.node.split('.').map(Number);

4if (major < 20) {

5 console.error(

dist/supabase.jsView file

Lines 1-21javascript

1import * as os from 'os';

2import { execSync } from 'child_process';

3import * as logger from './logger.js';

dist/sudosudo.jsView file

6* path) instead of letting the LLM interpret a privileged setup script. The L7: * installer URL is validated to be https://*.sudosudo.dev. This is how the L8: * Orquesta dashboard "activates" sudosudo on a box that already runs this agent — ... L10: */ L11: import { spawn } from 'child_process'; L12: const MARKER_RE = /<<ORQ_SUDOSUDO_INSTALL>>([\s\S]*?)<<\/ORQ_SUDOSUDO_INSTALL>>/; ... L17: try { L18: const spec = JSON.parse(m[1]); L19: if (typeof spec.install_url !== 'string' || !spec.install_url) ... L41: * Fetch + run the official sudosudo installer, streaming each output line to L42: * `onLine`. Resolves with the exit code. Rejects only if the URL is not a L43: * sudosudo.dev https URL or the process can't be spawned.

Critical

Download Execute

Source downloads or fetches remote code and executes it.

dist/sudosudo.jsView on unpkg · L6

•matchType = normalized_sha256 matchedPackage = orquesta-agent@0.2.246 matchedPath = dist/sudosudo.js matchedIdentity = npm:b3JxdWVzdGEtYWdlbnQ:0.2.246 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/sudosudo.jsView on unpkg

dist/init.jsView file

•Manifest-trigger-reachable source links an external AI-agent control path to a behavior-bearing write operation. token: options.token, apiurl, }; fs.writefilesync(configpath, json.stringify(config, null, 2) + '\n'); console.log(' created .orquesta.json'); const gitignorepath = path.join(cwd, '.gitignore'); const entry = '.orquesta.json'; if (fs.existssync(gitignorepath)) { const content = fs.readfilesync(gitignorepath, 'utf8'); if (!content.includes(entry)) { fs.appendfilesync(gitignorepath, `\n# orquesta hook config (contains token)\n${entry}\n`); console.log(' added .orquesta.json to .gitignore'); } } else { fs.writefilesync(gitignorepath, `# orquesta hook co

High

Trigger Reachable External Ai Agent Control Surface Mutation

Manifest-trigger-reachable source writes behavior-bearing configuration into a user or project AI-agent control surface.

dist/init.jsView on unpkg

•matchType = normalized_sha256 matchedPackage = orquesta-agent@0.2.246 matchedPath = dist/init.js matchedIdentity = npm:b3JxdWVzdGEtYWdlbnQ:0.2.246 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/init.jsView on unpkg

scripts/generate-manifest.shView file

•path = scripts/generate-manifest.sh kind = build_helper sizeBytes = 2081 magicHex = [redacted]

Medium

Ships Build Helper

Package ships non-JavaScript build or shell helper files.

scripts/generate-manifest.shView on unpkg

dist/hook.js#virtual:normalized:round1View file

•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 2

High

Semantic Analysis Limited

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

dist/hook.js#virtual:normalized:round1View on unpkg

dist/browser.jsView file

•matchType = normalized_sha256 matchedPackage = orquesta-agent@0.2.246 matchedPath = dist/browser.js matchedIdentity = npm:b3JxdWVzdGEtYWdlbnQ:0.2.246 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/browser.jsView on unpkg

scripts/build-binaries.shView file

•matchType = normalized_sha256 matchedPackage = orquesta-agent@0.2.246 matchedPath = scripts/build-binaries.sh matchedIdentity = npm:b3JxdWVzdGEtYWdlbnQ:0.2.246 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.

scripts/build-binaries.shView on unpkg

dist/executor.jsView file

•matchType = previous_version_dangerous_delta matchedPackage = orquesta-agent@0.2.251 matchedIdentity = npm:b3JxdWVzdGEtYWdlbnQ:0.2.251 similarity = 0.758 summary = stored previous version shares package body but lacks this dangerous source file

Critical

Previous Version Dangerous Delta

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

dist/executor.jsView on unpkg

Source & flagged code

22 flagged · loading source

package.jsonView file

•scripts.preinstall = node scripts/preinstall.js

High

Install Time Lifecycle Scripts

Package defines install-time lifecycle scripts.

package.jsonView on unpkg

•scripts.preinstall = node scripts/preinstall.js

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

•scripts.postinstall = node scripts/postinstall.js

Medium

Ambiguous Install Lifecycle Script

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

package.jsonView on unpkg

dist/ui/onboarding.jsView file

Findings

5 Critical14 High8 Medium5 Low

CriticalSame File Env Network Executiondist/index.js

CriticalHardcoded Runtime Data Exfiltrationdist/supabase.js

CriticalDownload Executedist/sudosudo.js

CriticalTrigger Reachable Dangerous Capabilitydist/index.js

CriticalPrevious Version Dangerous Deltadist/executor.js

HighInstall Time Lifecycle Scriptspackage.json

HighChild Processdist/ui/onboarding.js

HighShell

HighCredential Exfiltrationdist/index.js

HighTrigger Reachable External Ai Agent Control Surface Mutationdist/init.js

HighTrigger Reachable Credential Exfiltrationdist/index.js

High

Affected versions and remediation

This report applies to orquesta-agent@0.2.256.

See all versions of orquesta-agent for other recorded verdicts.

Review the evidence and your use of orquesta-agent@0.2.256 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: 2026-09-26 04:10:09.162Z (UTC).

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

orquesta-agent@0.2.256

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Affected versions and remediation

Runtime surface

Behavioral surface

Source & flagged code