Back to skill

Security audit

cargo-storage

Security checks for vulnerabilities and agentic risk

Overview

Review the webhook setup: it can expose a workspace access key with more permissions than needed.

Before using webhook setup, replace automatic first-token selection with an explicitly chosen token whose permissions you have checked. Transfer credentials privately and prefer an authorization header where supported. Revoke any privileged token already exposed through a shared URL. Verify the workspace and intended data changes, and use a reviewed CLI version; the external CLI and missing sibling security guidance were not inspected.

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
  • 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
91% confidence
Finding

The supplied commands manage model-associated execution runs, not model definitions, relationships, or business records. Listing runs also falls within the run-oriented functionality the description explicitly directs to cargo-orchestration. Although storage runs could support data imports, the supplied chunk shows only run creation and listing, without that business-data context.

Content

No source excerpt is available for this finding.

The webhook setup examples retrieve an arbitrary existing workspace API token and embed it in a shareable URL.

High
Category
Excessive Agency
Confidence
98% confidence
Finding

The examples retrieve an arbitrary existing workspace API token and package it into a URL intended for a third party. The guidance mentions zero-permission tokens and warns about secrets, but never verifies that the selected token has those restricted permissions; exposing a more privileged token could enable broader workspace access.

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
90% confidence
Finding

Although this is a CLI rather than an MCP server, the documented npx invocation can download and execute an unpinned release. A compromised upstream release could execute with the agent's local privileges and access Cargo credentials; the separately referenced version-pin guidance does not make this standalone command reproducible.

Content

No source excerpt is available for this finding.

The webhook setup prints an API-token-bearing URL without warning that it contains a credential.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The example selects the first existing workspace token without checking its permissions and prints it inside a webhook URL. This can expose a broadly privileged credential through tool output, conversation history, or downstream URL logs, even though building a webhook is a legitimate storage operation.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/examples/ingest-webhook.md (reported line 92)May include surrounding context.

bash
# one record
curl -X POST "$BASE/v1/models/$MODEL_UUID/records/ingest?token=$TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"email":"ada@example.com","company":"example.com"}'

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

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

The troubleshooting advice can install or execute whatever CLI release is currently resolved by npm rather than a reviewed version. Although the MCP label is inaccurate, this creates a supply-chain exposure because package installation and CLI execution occur with the agent's local privileges.

Content

No source excerpt is available for this finding.