Back to skill

Security audit

cargo-context

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently manages Cargo’s company knowledge base, with disclosed repository changes and human review of call-derived edits.

Confirm the workspace and repository audience before writing: edits and command-driven deletions push immediately. Review and redact sensitive call quotations and customer details before approving them. Prefer a reviewed, pinned CLI release and workspace-scoped credentials. This assessment covers the bundled instructions, not the external CLI implementation or missing sibling guidance.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk orchestrates company-data enrichment, an undeclared capability distinct from maintaining the described knowledge base. It does not read or update knowledge-base markdown, the sandbox, or the knowledge graph; its only file writes store enrichment results under /tmp. Although enrichment could inform later documentation, this chunk performs substantive data acquisition rather than merely implementing a declared knowledge-base operation.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx @cargo-ai/cli'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The bootstrap instructions invoke an unpinned npm CLI through npx, and the installation metadata explicitly selects @latest. Although this is a CLI rather than an MCP server, a compromised future release could execute code with the invoking user's privileges and access available credentials or company context; the reference to version-pinning guidance elsewhere does not secure this standalone installation path.

Content

No source excerpt is available for this finding.

The call-analysis workflow lacks a privacy warning about copying sales-call content into reusable workspace context.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow promotes sales-call material, including attributable quotes and customer facts, into reusable context that is immediately committed and consumed by downstream agents. It requires human approval for edits but provides no explicit privacy or audience check, creating a risk that restricted customer information or personal data reaches a broader audience and persists in repository history.

Content

No source excerpt is available for this finding.