Back to skill

Security audit

Perfex CRM (Perfex Connect)

Security checks for vulnerabilities and agentic risk

Overview

The skill provides scoped CRM access with read-only defaults and user-controlled changes, although approval can occur in chat rather than inside Perfex.

Start with a read-only token limited to the resources you need. Keep Ask Before Every Change enabled if each write must receive separate approval inside Perfex; otherwise, the documented fallback can commit an exact change after your agreement in chat.

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)

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 64)May include surrounding context.

md
3. **Writes go through approvals.** On most tokens a create or update does not write. The answer has status
   `pending_approval` and a `request_id`. Tell the user the change waits for approval in Perfex (Connect > Approvals),
   give the `review_url` if there is one, and check later with `connect_request_status`. Never say a change is done
   while it is pending. If an answer is a preview (no approval on this token), show it to the user and repeat the call
   with `confirm: true` only when the user asked for exactly that change.
4. **Respect refusals.** "read-only token", a missing permission, a resource you may not use, a rate limit (429) or
   "Record Changed" are final for this request. Explain it in one sentence and stop. Do not try another tool, another

The skill permits direct writes without the in-Perfex human approval promised by its manifest.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The description promises approval in Perfex, but this fallback explicitly permits committing changes with confirm: true when the token does not require that approval. Conversational user consent reduces the risk, and the recommended setup enables approvals, but neither preserves the promised in-Perfex review boundary when a differently configured token is used.

Content

No source excerpt is available for this finding.