Back to skill

Security audit

PDF.co

Security checks for vulnerabilities and agentic risk

Overview

The PDF.co integration makes sense, but setup can execute unverified downloaded code and introduce additional agent skills without clearly explaining those changes.

Review the setup before enabling this skill. Install a verified CLI release separately and disable startup skill synchronization with OO_SKILLS_SYNC_DISABLED=1 if you do not want additional agent skills installed or refreshed. Only submit documents authorized for OOMOL/PDF.co processing, and check credit costs for conversion actions. The remote installers remain unverified; the inspected artifact does not establish malicious intent.

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 (6)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied command executes an external installer rather than performing a PDF.co operation. Although installation can be a supporting implementation detail, the supplied content establishes no connection between the Oomol CLI and the declared PDF.co functionality. The downloaded script's contents and further effects are not visible.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The supplied code performs only a login operation, rather than handling PDF.co data. Authentication could be a supporting implementation detail, but this chunk provides no connection between the oo login and PDF.co. As supplied, its behavior does not substantiate the declared purpose.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The setup procedure pipes a mutable remote installer directly into a shell without pinning or independently verifying its contents; the PowerShell alternative has the same issue. Compromise of the installer distribution path could execute attacker-controlled code with the agent process's privileges, although the missing-CLI condition limits exposure and no malicious installer content is demonstrated.

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

The execution workflow sends document inputs to third-party services without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow instructs the agent to submit action payloads through OOMOL to PDF.co, which can include document URLs or raw HTML. Although the services are identified, the skill does not warn that supplied content or document links will be shared with external providers, which matters for confidential documents.

Content

No source excerpt is available for this finding.

The safety policy labels untagged PDF-generation actions as read-only, bypassing its write-confirmation requirement.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The blanket rule declaring untagged actions to be reads conflicts with untagged operations such as html_to_pdf, split_pdf, and url_to_pdf, which generate output artifacts and may consume account credits. This creates a concrete gap in the skill's own confirmation policy, although the listed actions do not indicate destructive modification of source documents.

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 83)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