Back to skill

Security audit

HoneyHive

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for managing HoneyHive, but its setup instructions include direct execution of unverified remote installer scripts.

Before installing, review or replace the oo CLI setup path with a safer version-pinned or signed installer flow. For normal HoneyHive use, pay close attention to write and destructive actions and only approve exact payloads and targets you intend to change.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:60
Finding
Unverified Remote Installation Scripts Are Downloaded and Executed Directly## Vulnerability Details **File Location**: `SKILL.md`, lines 60–64 **Vulnerability Type**: Remote payload retrieval and execution **Risk Level**: High The first-time setup instructions provide two platform-specific commands that immediately execute scripts retrieved from external URLs: ```bash curl -fsSL https://cli.oomol.com/install.sh | bash # macOS / Linux ``` ```powershell irm https://cli.oomol.com/install.ps1 | iex # Windows PowerShell ``` ### Technical Analysis Both installation methods pass mutable remote content directly to a command interpreter. The Unix command pipes the HTTP response into Bash, while the Windows command retrieves a PowerShell script and passes it to `Invoke-Expression`. The Skill does not pin a script version, verify a cryptographic signature or checksum, save the script for inspection, or constrain the commands that the installer may execute. HTTPS protects data in transit but does not establish that the current script content is identical to what was reviewed. Compromise of the hosting infrastructure, publishing account, DNS path, or upstream release pipeline could therefore turn these installation commands into arbitrary code execution. This behavior exceeds the minimum privilege required merely to operate the HoneyHive connector. The Skill's normal runtime permission is limited to `Bash(oo *)`, but the setup instructions introduce execution of unrestricted Bash or PowerShell code whose behavior is not represented in the audited project. ### Attack Path 1. The `oo` CLI is unavailable, and an agent or user follows the documented first-time setup. 2. The system retrieves `install.sh` or `install.ps1` from `cli.oomol.com`. 3. The remote script has been maliciously modified through compromise of the hosting server, publishing credentials, DNS infrastructure, or software distribution pipeline. 4. The response is passed directly to Bash or PowerShell without local review or integrity verification. 5. The remote pa ...[truncated 974 chars]
Remediation
## Remediation Suggestions 1. Replace pipe-to-shell and `Invoke-Expression` instructions with installation through a trusted, version-pinned package manager or signed release artifact. 2. Pin the CLI to a specific reviewed version rather than retrieving an unversioned installation script. 3. Publish SHA-256 checksums and, preferably, cryptographic signatures through an independently protected channel. 4. Download the artifact to a local file before execution, verify its signature or checksum, and abort on any mismatch. 5. Allow users to inspect the downloaded script before running it. For example, separate retrieval, verification, and execution into distinct commands. 6. Avoid requiring administrator or root privileges unless a documented installation step strictly requires them. 7. Document the files, directories, network destinations, and configuration changes performed by the installer. 8. Apply equivalent verification controls to both the Bash and PowerShell installation paths. 9. Prefer an installation flow resembling: ```bash curl -fL -o oo-installer.sh "https://trusted.example/releases/v1.0.2/install.sh" echo "<reviewed-sha256> oo-installer.sh" | sha256sum --check - less oo-installer.sh bash oo-installer.sh ``` The checksum must be obtained from a separately authenticated release channel, and a signed package is preferable to relying on a checksum alone.
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (2)

External Script Fetching

High
Category
Supply Chain
Content
- **`oo: command not found`** — install the oo CLI (other platforms: <https://cli.oomol.com/install-guide.md>):

  ```bash
  curl -fsSL https://cli.oomol.com/install.sh | bash    # macOS / Linux
  ```

  ```powershell
Confidence
95% confidence
Finding
The skill recommends installing software via a remote script piped directly into a shell (`curl ... | bash`), which executes unreviewed network content immediately. If the install endpoint, CDN, DNS, or transport path is compromised, this can lead to arbitrary code execution on the host running the skill.

Vague Triggers

Medium
Confidence
96% confidence
Finding
The skill description is overly broad: it instructs the agent to use this skill for ANY HoneyHive request, including reads, writes, and deletions. Broad routing increases the chance the skill is invoked in situations where safer, narrower handling or additional review should apply, especially because the same skill exposes destructive operations.

Static analysis

No suspicious patterns detected.