Back to skill

Security audit

withhuman-cli

Security checks for vulnerabilities and agentic risk

Overview

The skill clearly describes a user-controlled approval integration, with ordinary installer and data-sharing risks to consider.

Trust the download source before running its installer. Confirm the server, organization, runtime, and gateway permissions, and understand that tool-call details are sent for approval. Check which rules approve automatically. When disconnecting permanently, remove the adapter and revoke its instance in the app.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

External Script Fetching

High
Category
Supply Chain
Confidence
96% confidence
Finding

The onboarding command immediately executes a remotely fetched, unpinned shell script without verifying its integrity or authenticity beyond HTTPS. Compromise of the download infrastructure could execute arbitrary commands with the invoking user's privileges, potentially affecting credentials and runtime approval hooks; interactive onboarding does not verify the downloaded code.

Content

Scanner excerpt · SKILL.md (reported line 51)May include surrounding context.

one-time setup code, so the person must copy it from their own page:

sh
curl -fsSL https://downloads.withhuman.ai/install.sh | sh -s -- onboard --url https://app.withhuman.ai --code <code>

Leave the command running while the person finishes the browser steps

External Script Fetching

High
Category
Supply Chain
Confidence
96% confidence
Finding

The installation command pipes mutable remote content directly into a shell, giving the download endpoint control over code executed locally. This is a supply-chain exposure rather than evidence of malicious intent, but installing software that manages runtime hooks makes a compromised installer particularly consequential.

Content

Scanner excerpt · SKILL.md (reported line 64)May include surrounding context.

Install the CLI (skip if withhuman version already works):

sh
curl -fsSL https://downloads.withhuman.ai/install.sh | sh

Then connect a runtime:

Chaining Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool calls are chained to bypass individual safety checks or escalate capabilities beyond what any single tool call would allow.

Content

Scanner excerpt · SKILL.md (reported line 64)May include surrounding context.

Install the CLI (skip if withhuman version already works):

sh
curl -fsSL https://downloads.withhuman.ai/install.sh | sh

Then connect a runtime: