Back to skill

Security audit

Accounts Payable Agent

Security checks for vulnerabilities and agentic risk

Overview

This accounts-payable skill is coherent but asks an agent to autonomously move real funds across high-risk payment rails without enough scoping or confirmation safeguards.

Review this skill carefully before installing. It should only be used with a payment environment that enforces human approval, authenticated callers, vendor allowlists, immutable idempotency keys, cumulative limits, audit logs, and reconciliation before any retry or rail failover. Do not give it unrestricted access to live payment tools or crypto, stablecoin, wire, payroll, or vendor-payment accounts.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
AGENTS.md:93
Finding

Unauthenticated Inter-Agent Requests Can Directly Trigger Payments

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
AGENTS.md:69
Finding

Recurring Payment Workflow Omits Required Idempotency Enforcement

Content
View full analysis
SPEND_LIMIT) { await escalate(bill, "Exceeds autonomous spend limit"); continue; } const result = await payments.send({ to: bill.recipient, amount: bill.amount, currency: bill.currency, reference: bill.invoiceId, memo: bill.description }); await logPayment(bill, result); await notifyRequester(bill.requestedBy, result); } ``` ### Technical Analysis The recurring-payment loop sends each returned bill without first checking whether its invoice has already been paid or atomically reserving its idempotency key. This conflicts with the stated requirement to perform an idempotency check before every transaction. Logging occurs only after `payments.send`. If the process terminates, times out, or fails between sending and logging, the next execution may process the same bill again. Duplicate scheduler records, concurrent workers, stale bill status, or deliberate replay can produce the same result. A separate check followed by a send would also remain vulnerable to a time-of-check/time-of-use race unless the reservation and state transition are atomic or the underlying payment provider enforces the same idempotency key. ### Attack Path 1. A scheduled bill remains marked as due, is duplicated in the scheduler, or is deliberately replayed. 2. The bill amount remains below `SPEND_LIMIT`. 3. The workflow does not query prior payment state or reserve `bill.invoiceId`. 4. `payments.send` submits the transfer. 5. The workflow is executed again, or a concurrent worker processes the same bill. 6. The same obligation is paid a second time because no durable atomic uniqueness control prevents the repeated send. ...[truncated 724 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SOUL.md:15
Finding

Automatic Payment-Rail Failover Can Duplicate Ambiguous Transactions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (6)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs autonomous execution of vendor and contractor payments across real payment rails, including wire, crypto, and stablecoins, but provides no explicit warning that actions may move irreversible real funds. In an agent ecosystem, this omission increases the chance that a caller or orchestrator treats payment execution as routine automation and triggers financial transfers without sufficient human confirmation, especially for irreversible rails like crypto and wires.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

A skill that presents itself as moving money should warn that actions may be financial, high impact, and potentially irreversible, especially across crypto or stablecoin rails. Without explicit warnings and confirmation expectations, users may invoke the skill casually or agents may route sensitive payment requests through it without adequate human review.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The identity text advertises broad authority to 'move money across any rail' while the skill metadata says it is an accounts-payable agent. That mismatch can mislead users or downstream orchestrators into granting or attempting broader payment authority than intended, increasing the chance of unauthorized or unsafe financial actions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Claiming support for crypto and stablecoin transfers expands the implied financial surface into higher-risk and often irreversible payment rails without any justification or safeguards in the description. This can prompt use in contexts involving irreversible transfers, compliance-sensitive assets, or unsupported transaction types based solely on misleading documentation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation guidance is extremely broad: 'Reference this agent by name or specialty' can cause the skill to be invoked from vague mentions rather than explicit user intent. In a finance-oriented agent that can process payments, unintended activation increases the risk of the model adopting payment-execution behavior in the wrong context, potentially leading to unauthorized payment workflows or unsafe tool use.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% 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 · SOUL.md (reported line 18)May include surrounding context.

md
### Error Handling
- If a payment rail fails, try the next available rail before escalating
- If all rails fail, hold the payment and alert — do not drop it silently
- If the invoice amount doesn't match the PO, flag it — do not auto-approve

## 💭 Your Communication Style
- **Precise amounts**: Always state exact figures — "$850.00 via ACH", never "the payment"