Back to skill

Security audit

Baidu Qianfan

Security checks for vulnerabilities and agentic risk

Overview

The skill needs review because setup executes unverified downloaded software and its safety rules misclassify some potentially chargeable AI requests as reads.

Review and verify the CLI separately before allowing installation, including its automatic installation and synchronization of other agent skills. Require clear approval for submitted content and potential generation charges. Remote installers and hosted connector implementations were not verified; this review found concrete control gaps, not evidence of deliberate malicious behavior.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (5)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk installs software from Oomol rather than performing Baidu Qianfan operations. Nothing in the chunk establishes that this installation supports the declared Qianfan functionality. The downloaded script's contents are unavailable, so its further behavior cannot be assessed.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

Piping a mutable HTTPS download directly into Bash executes whatever code the endpoint returns with the invoking user's privileges, without integrity verification or review. A compromised distribution endpoint could therefore compromise local files and credentials; the legitimate OOMOL dependency context does not remove this supply-chain risk.

Content

Scanner excerpt · SKILL.md (reported line 79)May include surrounding context.

The safety policy labels all untagged actions as reads, but untagged actions include chat and AI-search completion generation.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The blanket statement that untagged actions are reads incorrectly includes create_chat_completion and create_ai_search_completion, which submit inputs for inference and can incur charges. This can bypass the skill's confirmation requirement for the exact payload and effect, although the actions remain legitimate when explicitly requested by the user.

Content

No source excerpt is available for this finding.

The setup workflow executes downloaded installer scripts without warning the user about remote code execution or local system changes.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

A missing CLI triggers instructions to execute downloaded installation scripts without explicit approval for local code execution or system changes. This is a legitimate dependency setup step, but it crosses from remote service operations into local software installation; the missing-command condition limits when it occurs, not what the installer can do.

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 100)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