Back to skill

Security audit

PayLock SOL Escrow

Security checks for vulnerabilities and agentic risk

Overview

Review recommended: this escrow skill handles financial actions and tokens, but its custody and security claims do not consistently match the bundled client behavior.

Install only after reviewing the trust model. Treat production v1 as custodial, use only localhost or a trusted HTTPS PayLock endpoint, avoid passing payer/payee tokens on the command line, and verify token scope, expiry, revocation, server custody, and release behavior before using real funds or sensitive deliverables.

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
scripts/paylock.py:27
Finding

Authentication Tokens Exposed Through Command-Line Arguments

Content
View full analysis
/cmdline` on Linux - Process-monitoring and orchestration systems - Audit, debugging, and terminal-session logs - Wrapper scripts or automation logs This behavior also contradicts `SKILL.md:21-25`, which states that authentication tokens are passed through environment variables and never through CLI arguments. Although the dedicated `deliver_contract.py` and `verify_contract.py` scripts support environment variables, the documented unified `paylock.py` interface does not. ### Attack Path 1. A user invokes the unified CLI with `--payee-token` or `--payer-token`. 2. The token ...[truncated 1073 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/paylock_api.py:14
Finding

Sensitive Financial Requests May Be Transmitted over Plaintext HTTP

Content
View full analysis
None: self.base_url = ( base_url or os.getenv("PAYLOCK_API_BASE") or DEFAULT_API ).rstrip("/") def request( self, method: str, path: str, payload: Optional[Dict[str, Any]] = None, timeout: int = 30, ) -> Dict[str, Any]: url = f"{self.base_url}/{path.lstrip('/')}" data = None headers = {"Accept": "application/json"} if payload is not None: data = json.dumps(payload).encode("utf-8") headers["Content-Type"] = "application/json" req = urllib.request.Request( url=url, method=method.upper(), data=data, headers=headers, ) try: with urllib.request.urlopen(req, timeout=timeout) as resp: ``` The corresponding documentation permits an operator-provided endpoint without requiring HTTPS: ```bash export PAYLOCK_API_BASE="http://localhost:8767" ``` ```text Agents running their own PayLock instance use localhost. For hosted PayLock, set the URL provided by your PayLock operator. ``` ### Technical Analysis Using plaintext HTTP for a loopback-only development service can be acceptable under a defined local trust model. However, the client accepts arbitrary remote HTTP endpoints through both `PAYLOCK_API_BASE` and the `--api` argument without validation or warning. The request body can contain: - Payer and payee authentication tokens - Wallet addresses - Transaction hashes - Contract descriptions and participant identifiers - Delivery payloads and hashes - Verification requests that may cause funds to be released When a rem ...[truncated 1811 chars]
Remediation
View remediation

other

Warning
Location
SKILL.md:1
Finding

Security and Custody Claims Conflict with the Bundled Client Behavior

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (5)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes the skill as non-custodial, but later documentation states that v1 production is custodial escrow. This is security-relevant misrepresentation because users and agent operators may make materially different trust and fund-risk decisions based on whether the system ever takes custody of assets.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documentation repeatedly markets the system as non-custodial while later disclosing that the production architecture is custodial. In a financial escrow skill, that inconsistency is especially dangerous because it can mislead operators into exposing funds to a custody model they did not intend to trust.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill exposes capabilities that rely on environment variables and network access, but the manifest does not declare any tool scope or permissions. In an agent ecosystem, this weakens the trust boundary because a user or platform cannot accurately understand what resources the skill expects to access before execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The CLI forwards highly sensitive values such as payer/payee tokens and delivery payloads directly to a remote API, yet this file provides no warning, consent prompt, or disclosure to the operator that secrets and potentially proprietary work product are being transmitted off-host. In the context of an AI-agent escrow skill that advertises 'No browser needed' and is likely to be invoked non-interactively, this increases the risk of accidental secret exfiltration or unintended disclosure of sensitive deliverables to a configurable endpoint.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The client performs network requests with urlopen, and when payload is provided it serializes and sends JSON to the configured API endpoint. Although the module docstring states it is an API client, there is no user-facing warning, confirmation, or inline disclosure near the request path about transmitting data over the network.

Content

No source excerpt is available for this finding.