Back to skill

Security audit

Allowance

Security checks for vulnerabilities and agentic risk

Overview

This is a legitimate-looking purchase-wallet skill, but it gives an agent real-money checkout authority and installs mutable external code globally with limited user control.

Review carefully before installing. Use this only when you want an agent to complete purchases, prefer pinned and verified versions of both the skill and CLI, and do not let it proceed on vague payment or booking requests without a concrete merchant, item, total cap, and shipping expectations.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:26
Finding

Unpinned Global npm Package Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 26-30; README.md, lines 19-23
Vulnerability Type: Unpinned and unaudited third-party dependency installation
Risk Level: Medium

Vulnerable Code

SKILL.md, lines 26-30:

markdown
If `allowance` is not on PATH, install it globally first:

npm install -g @allowance/cli

text

README.md, lines 19-23:

markdown
Install the Allowance CLI:

```bash
npm install -g @allowance/cli
text

### Technical Analysis

The documented command installs the latest available version of `@allowance/cli` globally without pinning an exact version or verifying package integrity. The dependency's source code is not included in the audited project, so its behavior and npm lifecycle scripts could not be reviewed.

A global npm installation can execute package lifecycle scripts and create or replace executable files in a host-wide npm binary location. Consequently, the effective code executed by this skill can change after the skill itself has been audited. Compromise of the npm publisher account, package registry, release pipeline, or an unexpectedly malicious future release could expose the agent environment to arbitrary code execution with the permissions of the account running npm.

### Attack Path

1. An attacker compromises the npm publisher account, package release pipeline, or another component responsible for publishing `@allowance/cli`.
2. The attacker publishes a malicious new version containing a lifecycle script or modified CLI executable.
3. An agent follows the skill instructions and runs `npm install -g @allowance/cli`.
4. npm retrieves the latest compromised version because no version or integrity value is pinned.
5. Malicious installation logic executes with the privileges of the agent user.
6. The installed CLI may then access files available to that user, capture purchase-related data, alter commands, or communicate with at
...[truncated 646 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the CLI to a reviewed exact version, such as @allowance/cli@X.Y.Z, rather than installing the latest release.
  2. Lock and verify the expected package integrity hash before installation.
  3. Publish and verify package provenance or cryptographic release signatures.
  4. Prefer a project-local, isolated, or sandboxed installation rather than a global installation.
  5. Disable npm lifecycle scripts where operationally possible, or explicitly audit all lifecycle scripts before allowing them to execute.
  6. Run the CLI under a dedicated least-privileged account with restricted filesystem, environment, and network access.
  7. Document a controlled upgrade process requiring review and integrity verification before changing the pinned version.

T08 · Insecure Dependencies

Warning
Location
README.md:13
Finding

Skill Installation from a Mutable Git Branch

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 13-17
Vulnerability Type: Mutable external dependency reference
Risk Level: Medium

Vulnerable Code

markdown
OpenClaw can install this skill from GitHub:

```bash
openclaw skills install git:useallowance/allowance-skill@main
text

### Technical Analysis

The installation command references the mutable `main` branch instead of an immutable commit hash or a verified release artifact. The contents retrieved by this command may therefore change after the current artifact has been audited.

Since a skill consists of instructions that influence agent behavior, an attacker who can modify the upstream branch could introduce new directives affecting purchases, sensitive checkout data, tool usage, or external communication. Users executing the documented command would receive those modified instructions without assurance that they match the reviewed version.

### Attack Path

1. An attacker compromises the upstream GitHub repository, a maintainer account, or the repository's update pipeline.
2. The attacker pushes modified skill instructions to the `main` branch.
3. A user follows the README and installs the skill using the mutable branch reference.
4. OpenClaw retrieves and activates the attacker-modified skill content.
5. The agent follows the altered instructions during purchase workflows.
6. Depending on the injected instructions and available tools, the attacker could attempt to redirect transactions, solicit sensitive information, or cause unsafe command execution.

### Impact Assessment

Exploitation could alter the behavior of agents that install or update the skill after the upstream branch is compromised. The practical scope depends on the permissions and tools available to the affected agent, potentially including command execution, browser interaction, and access to purchase-related information.

No malicious branch content was identi
...[truncated 143 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the mutable @main reference with an immutable, reviewed commit SHA.
  2. Prefer signed, versioned release tags and verify the tag signature before installation.
  3. Publish checksums for release artifacts and require checksum verification during installation.
  4. Configure update processes so new skill versions are reviewed before activation.
  5. Protect the upstream repository with mandatory multi-factor authentication, branch protection, signed commits, and required code review.
  6. Clearly document the exact audited version so users can confirm that installed content matches the reviewed artifact.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill uses very broad activation language such as 'buy, order, book, reserve, pay, or spend money,' which can cause the agent to invoke a spending-capable workflow for loosely related user requests. In a purchase wallet context, unintended invocation is more dangerous than usual because it can lead to unnecessary payment flows, premature checkout actions, or confusing escalation into financial operations.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill description uses very broad spending-related triggers like 'buy something,' 'pay,' or 'spend money on their behalf,' which can cause the skill to be invoked for ambiguous requests that are not intended to authorize a purchase workflow. In a payments skill, misrouting is especially dangerous because invocation can initiate wallet setup, approval requests, or downstream purchasing steps in contexts where the user's intent was exploratory rather than transactional.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The routing rule instructs the agent to use the Allowance CLI by default whenever the user asks to 'buy, order, book, reserve, pay, or spend money,' without clear exclusions or a requirement to first verify that the user actually wants this wallet workflow. Because this is a real-money purchasing skill, ambiguous auto-routing increases the chance of unintended invocation and can push the agent into collecting sensitive information or creating approval requests for the wrong task.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 · SKILL.md (reported line 15)May include surrounding context.

md
### Routing rule

Use the Allowance CLI by default. When the user asks to buy, order, book,
reserve, pay, or spend money, follow the CLI flow below without asking whether
to use CLI or MCP.

Only use the MCP integration if the user explicitly asks for Allowance MCP,

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

This line explicitly authorizes the agent to complete checkout end-to-end, including submitting the final order, without a second confirmation once an approval request is accepted. In a purchasing context, that is highly sensitive autonomous action: if the approval request was overly broad, misunderstood, socially engineered, or mismatched to checkout contents, the agent can finalize a real monetary transaction and transmit shipping/contact data without an immediate final human review.

Content

Scanner excerpt · SKILL.md (reported line 147)May include surrounding context.

md
Approval on the user's iPhone is the **only human gate**. Once `allowance request`
   returns `approved`, complete checkout end-to-end — including clicking submit /
   place order — without asking the user to confirm again. Re-prompt the user only if:
   - the realized total would exceed the approved cap (`--amount`), or
   - you must substitute a materially different item or variant than the approval
     reason allows, or

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% 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 · SKILL.md (reported line 248)May include surrounding context.

text

The hosted MCP is an alternate integration path, not the default purchase flow.
Do not ask the user to choose between CLI and MCP. If the user did not ask for
MCP, use the Allowance CLI.