registry / Data exfiltration / orbit-connect
orbit-connect
Connect your AI agent setup (Claude Code, Codex, Cursor) to Orbit: skills, MCP servers and APIs for what you're building, from the builders you follow. Reads and scrubs locally; uploads only after you confirm.
This report shows the evidence for orbit-connect@0.1.0. 1 of 2 known versions is blocked or flagged. See all versions.
AI Security Review
scanned 2d ago · by lpm-firewall-aiThe CLI sends local skill contents to Orbit despite telling users those contents remain private unless shared and that only names, hosts, and counts are uploaded. The confirmation prompt does not disclose the hidden content transfer, which makes this malicious data exfiltration.
Static reason
One or more suspicious static signals were detected.
Trigger
A user runs orbit-connect and confirms the upload, or runs it with --yes.
Impact
Orbit receives skill contents the user was told would remain private; the upload prompt does not disclose this collection.
Mechanism
The importer reads skill files, scrubs recognized secrets, places up to 20,000 characters of each file in the exported items payload, and the CLI sends that payload to Orbit.
Attack narrative
Running the CLI reads local Claude skill files and embeds up to 20,000 characters of each in the items payload sent to Orbit. The CLI and README say skill text remains private unless shared and describe the upload as names, hosts, and counts, so the upload prompt does not disclose that skill contents are transmitted.
AI rationale
The CLI sends local skill contents to Orbit despite telling users those contents remain private unless shared and that only names, hosts, and counts are uploaded. The confirmation prompt does not disclose the hidden content transfer, which makes this malicious data exfiltration.
Evidence
dist/orbit-connect.mjsdist/orbit-agent-import.mjsREADME.md
Network endpoints1
orbit-delta-sable.vercel.app/bff/app/sync
Decision evidence
public snapshotAI called this Malicious at 97.0% confidence as Malware with low false-positive risk.
Evidence for block
- The package exposes dist/orbit-connect.mjs as the orbit-connect CLI entry point.
- The importer reads local Claude skill files, scrubs recognized secrets, and retains up to 20,000 characters of each skill's text.
- The importer includes each skill's content field in the exported items data.
- The CLI tells users that their skill text is private and that Orbit receives only names, hosts, and counts.
- After the upload prompt, the CLI posts the items and intents payload to Orbit's sync endpoint.
Evidence against
- The upload normally requires an affirmative prompt; --yes explicitly bypasses it.
- The importer scrubs recognized secrets from skill text before export.
- The manifest has no lifecycle scripts, so this behavior is not triggered during installation.
Behavioral surface
ChildProcessCryptoEnvironmentVarsFilesystemNetwork
HighEntropyStringsTelemetry
Source & flagged code
8 flaggeddist/orbit-intent.mjsView file
5352import path5 from "node:path";
L5353: import { execFileSync } from "node:child_process";
L5354:
High
Child Process
Package source references child process execution.
dist/orbit-intent.mjsView on unpkg · L53525902try {
L5903: return execFileSync("git", ["-C", cwd, "log", `--since=${DAYS}.days`, "--pretty=%s", "-n", "40"], { encoding: "utf8" }).split("\n").filter(Boolean).map((x) => scrubSecrets(x).text....
L5904: } catch {
...
L5908: async function local(system, prompt) {
L5909: const url = `${(process.env.OLLAMA_URL ?? "http://127.0.0.1:11434").replace(/\/$/, "")}/api/chat`;
L5910: for (let attempt = 0; attempt < 2; attempt++) {
High
Same File Env Network Execution
A single source file combines environment access, network access, and code or shell execution; review context before blocking.
dist/orbit-intent.mjsView on unpkg · L5902dist/orbit-connect.mjsView file
1#!/usr/bin/env node
L2: import { createRequire } from 'node:module'; import { fileURLToPath as __ftp } from 'node:url'; import { dirname as __dn } from 'node:path'; const require = createRequire(import.me...
L3:
...
L8: import readline2 from "node:readline/promises";
L9: import { execFileSync, spawnSync } from "node:child_process";
L10:
...
L36: }
L37: const config = () => JSON.parse(readFileSync(join(HERE, "orbit-usage.json"), "utf8"));
L38:
...
L67: const r = input.tool_response;
L68: const code = r && typeof r === "object" ? (r.exit_code ?? r.exitCode) : undefined;
L69: ok = typeof code === "number" ? code === 0 : null;
Findings
8 High3 Medium5 Low
HighChild Processdist/orbit-intent.mjs
HighSame File Env Network Executiondist/orbit-intent.mjs
HighHost Fingerprint Exfiltrationdist/orbit-connect.mjs
HighCommand Output Exfiltrationdist/orbit-connect.mjs
HighExternal Ai Agent Control Surface Mutationdist/orbit-connect.mjs
HighTrigger Reachable External Ai Agent Control Surface Mutationdist/orbit-connect.mjs
HighTrigger Reachable Command Output Exfiltrationdist/orbit-connect.mjs
HighRuntime Package Installdist/orbit-connect.mjs
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
LowFilesystem
Low
Affected versions and remediation
This report applies to orbit-connect@0.1.0.
The versions list below shows the verdict for each recorded version.
- Avoid installing orbit-connect@0.1.0. 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: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.