osborn@0.9.181
Voice AI coding assistant - local agent that connects to Osborn frontend
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThe package contains a serious exposed-runtime data and agent-storage mutation vulnerability, but source inspection does not establish malicious install-time behavior or a hidden attacker endpoint. It should be warned on rather than blocked as malware.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
Run the osborn CLI and allow a network client to reach its API port.
Impact
Sensitive Claude conversation, tool-output, and skill files can be disclosed or attacker-controlled files can be written into Claude storage.
Mechanism
Unauthenticated session archive export and import over a broadly bound HTTP server.
AI rationale
The package contains a serious exposed-runtime data and agent-storage mutation vulnerability, but source inspection does not establish malicious install-time behavior or a hidden attacker endpoint. It should be warned on rather than blocked as malware.
Evidence
dist/index.js~/.claude/projects~/.claude/skills
Network endpoints2
0.0.0.0code-server.dev/install.sh
Decision evidence
public snapshotAI called this Suspicious at 93.0% confidence as Critical Vulnerability with low false-positive risk.
Evidence for warning
- The runtime starts an HTTP server on all interfaces by default with permissive cross-origin access.
- The export route sends the complete Claude projects and skills directories without authentication unless an optional environment variable is configured.
- The import route accepts archives and merges their contents into the Claude projects and skills directories under the same optional-auth model.
- A frontend action can download and execute a remote code-server installer through a shell pipeline.
Evidence against
- package.json has no install, postinstall, or preinstall lifecycle hook; its only lifecycle hook is prepublishOnly.
- The observed OpenAI and OpenRouter requests are feature-aligned model and speech service calls, not evidence of hidden credential theft.
- The risky behavior occurs when the CLI runtime is started, not during package installation.
Behavioral surface
ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetwork
Source & flagged code
9 flagged · loading sourcebin/cli.jsView file
2L3: import { spawn } from 'child_process'
L4: import { fileURLToPath } from 'url'
High
dist/jsonl-search.jsView file
34// @vscode/ripgrep bundles the rg binary — this is our own dependency
L35: const { rgPath } = require('@vscode/ripgrep');
L36: if (rgPath && existsSync(rgPath))
Medium
Dynamic Require
Package source references dynamic require/import behavior.
dist/jsonl-search.jsView on unpkg · L34dist/index.jsView file
2455const which = await new Promise((res) => {
L2456: const p = spawn('which', ['code-server']);
L2457: let out = '';
...
L2464: await new Promise((res, rej) => {
L2465: const install = spawn('sh', ['-c', 'curl -fsSL https://code-server.dev/install.sh | sh'], {
L2466: env: { ...process.env },
L2467: stdio: ['ignore', 'pipe', 'pipe'],
Findings
5 Critical3 High5 Medium5 Low
CriticalSame File Env Network Executiondist/index.js
CriticalCredential Exfiltrationdist/index.js
CriticalCommand Output Exfiltrationdist/index.js
CriticalDownload Executedist/index.js
CriticalPrevious Version Dangerous Deltadist/index.js
HighChild Processbin/cli.js
HighShell
HighSemantic Analysis Limiteddist/session-access.js#virtual:normalized:round1
MediumDynamic Requiredist/jsonl-search.js
MediumNetwork
MediumEnvironment Vars
MediumShips Build Helperdeploy.sh
MediumStructural Risk Force Deep Review
Affected versions and remediation
This report applies to osborn@0.9.181.
See all versions of osborn for other recorded verdicts.
- Review the evidence and your use of osborn@0.9.181 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: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.