Back to skill

Security audit

GigaMail

Security checks for vulnerabilities and agentic risk

Overview

The skill’s email, calendar, optional file access, and automation features are disclosed and include clear user approval controls.

Connect only accounts and knowledge files you want the agent to read. Read content may enter your model provider’s context. Review action previews, enable automatic replies only with suitable limits, and verify approval behavior in the server version you install.

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
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Identity configuration could support email setup, but adding a local pricing spreadsheet to an identity is not explained by the declared email and calendar workflows. The chunk does not establish whether the file is transmitted or whether approval controls are bypassed; the mismatch is the undisclosed local-file operation.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 298)May include surrounding context.

md
the server's `env` block (see Setup step 3).
- Approvals the user grants "don't do anything" → same cause: server and
  console must share `GIGAMAIL_ROOT`.
- Approving from the CLI fails with "no consent backend" → that machine
  has no Windows Hello / Touch ID. On Linux the user sets a PIN with
  `gigamail approvals pin`; elsewhere they approve from the desktop
  console. `gigamail approvals check` shows which verification the machine

Static analysis

No suspicious patterns detected.