Back to skill

Security audit

LiteLLM

Security checks for vulnerabilities and agentic risk

Overview

Review is needed because the skill treats potentially billable AI requests as safe reads and understates the effects of its CLI setup.

Before enabling this skill, require user-directed generation with agreed inputs, model, and spending limits. Use a restricted LiteLLM key. Review the CLI installer and its managed-skill synchronization behavior, disabling synchronization if unwanted. Only SKILL.md is bundled; installer contents and live connector behavior were not verified.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command performs remote script execution rather than searching or reading LiteLLM data. Nothing in the chunk establishes this installer as a supporting implementation detail for the declared purpose. The downloaded script's contents and downstream effects are not shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied code performs only a login operation, not LiteLLM searching or reading. Authentication could support an accurately described skill, but this chunk provides no evidence connecting the oo login to LiteLLM or implementing the declared functionality.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The installation command immediately executes a remotely fetched, mutable script without checking a pinned version, signature, or independently trusted digest; the PowerShell alternative has the same problem. Compromise of the distribution endpoint could execute arbitrary code with the agent process's privileges, exposing local files and credentials, although the missing-CLI condition limits when this path is reached.

Content

Scanner excerpt · SKILL.md (reported line 60)May include surrounding context.

The skill exposes model inference operations despite describing its scope as searching and reading data.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest at L03 describes searching and reading data, but L40–L42 expose chat completion, embedding generation, and message generation. These operations submit inputs to model providers and can consume paid inference budgets rather than merely retrieve existing LiteLLM data.

Content

No source excerpt is available for this finding.

The safety instructions incorrectly classify untagged inference actions as reads safe to run directly.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

All three inference actions are untagged, so the blanket rule labels them reads that are safe to run directly even though they transmit payloads to upstream providers and can incur charges. This can encourage inference outside the user's intended read-only scope, although explicit user requests for inference and LiteLLM budget controls reduce the risk.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 81)May include surrounding context.

md
- **`OOMOL_INSUFFICIENT_CREDIT`** — OOMOL billing stop. Recharge at `https://console.oomol.com/billing/token-recharge` before retrying.

- **HTTP 402 without `OOMOL_INSUFFICIENT_CREDIT`** — the third-party account itself has no active plan or credit. Report the provider error message as is; do not ask the user to recharge OOMOL.

## Resources