Back to skill

Security audit

cargo-billing

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this billing-reporting skill suspicious because scanners report that it can launch a live workflow during cost estimates, beyond the stated reporting purpose.

Consider the scanner's do-not-install recommendation before installing. The findings point to live workflow side effects, an unpinned CLI package, and unsafe handling guidance for payment details. Coverage is incomplete: although all five bundled components were scanned, missing local references prevented a complete analysis, so the results do not establish that unreviewed material is safe.

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 (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The estimation procedure expands a billing inquiry into a live workflow execution without requiring separate authorization. Even a single-record run can spend credits or trigger external writes, messages, or other workflow actions that the user did not authorize merely by asking about cost.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The supplied commands enumerate resources but do not retrieve credits, usage, costs, plans, invoices, or payment details. Looking up workflows, agents, and connectors could support spend reporting, but the chunk contains only inventory operations, including storage-model enumeration with no evident billing connection. Its demonstrated purpose therefore materially differs from the declared purpose.

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
97% confidence
Finding

The bootstrap recommends executing an unpinned npm package, although the detector incorrectly calls it an MCP server. A compromised or malicious future package release could execute code with the agent's local privileges and access billing credentials; the reference to a version pin elsewhere does not constrain these standalone commands.

Content

No source excerpt is available for this finding.

The cost-estimation procedure creates a live workflow run rather than only reporting billing information.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The cost-estimation steps instruct the agent to create an actual workflow run rather than merely inspect existing usage. There is no dry-run restriction or approval gate, so a cost question can lead to billable activity and unintended changes in connected systems.

Content

No source excerpt is available for this finding.

The cost-estimation workflow instructs the agent to execute a live workflow without warning about its non-billing side effects.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The single-record sampling instruction omits warnings that the workflow may contact people, modify records, or invoke other external actions. Limiting the record count reduces scale but does not make those actions safe or authorized in a billing-reporting context.

Content

No source excerpt is available for this finding.

The card-stdin example contradicts its claim that the card number stays out of shell history.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Piping a literal JSON object through echo does not keep its card number and CVC out of shell history or agent tool logs. The example uses test data, but its explicit claim encourages the same unsafe handling of real payment credentials.

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
97% confidence
Finding

The recovery instructions recommend installing or executing the current npm release without an exact version constraint; this is a CLI dependency, not an MCP server. A compromised release could execute with the agent's privileges and expose local credentials, making automatic installation particularly sensitive in a billing workflow.

Content

No source excerpt is available for this finding.