Back to skill

Security audit

Openclaw Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is coherent, but it can use an Anycast token to query external systems, message remote agents, and write tenant-shared memory without clear safeguards.

Install only if you trust the Anycast portal URL and the token scope. Use a least-privilege, revocable token; keep ANYCAST_PORTAL_URL pinned to a trusted HTTPS origin; require explicit user approval before connector queries, remote-agent messages, or tenant memory writes; and avoid storing secrets or untrusted instructions in shared memory.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:17
Finding

Configurable API Endpoint Can Receive the Agent Token

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 17-24
Vulnerability Type: Credential disclosure through an unrestricted configurable endpoint
Risk Level: Medium

Vulnerable Code

markdown
You can interact with the Anycast agent platform using HTTP REST calls. All requests require the `x-agent-token` header set to `$ANYCAST_API_TOKEN`. The base URL is `$ANYCAST_PORTAL_URL` (default: `https://agents.anycast.com`).

## List Online Agents

```bash
curl -s -H "x-agent-token: $ANYCAST_API_TOKEN" \
  "$ANYCAST_PORTAL_URL/api/agents?limit=20" | jq '.agents[] | {name, status, lastSeenAt}'
text

The same token-and-configurable-URL pattern is repeated by the other request examples in `SKILL.md`.

### Technical Analysis

The skill directs the agent to include `ANYCAST_API_TOKEN` in the `x-agent-token` header while deriving the request destination entirely from `ANYCAST_PORTAL_URL`. Although the legitimate default is documented as `https://agents.anycast.com`, the instructions do not require HTTPS, validate the destination host, restrict redirects, or require confirmation before forwarding the credential to a custom origin.

Consequently, an attacker who can influence the environment or configuration can set `ANYCAST_PORTAL_URL` to an attacker-controlled endpoint. When a documented request is invoked, `curl` sends the token to that endpoint. The attacker can then attempt to replay the captured token against the legitimate Anycast API. The actual access available depends on the token's server-side authorization scope.

### Attack Path

1. The attacker gains the ability to modify or influence `ANYCAST_PORTAL_URL`, such as through a poisoned environment, unsafe deployment configuration, or misleading setup instructions.
2. The attacker changes the variable to an endpoint under their control, potentially using plain HTTP.
3. A user or agent follows one of the documented `curl` workflows.
4. The request tra
...[truncated 1062 chars]
Remediation
View remediation

Remediation Suggestions

  1. Restrict the default configuration to the canonical https://agents.anycast.com origin.
  2. Validate ANYCAST_PORTAL_URL before every credential-bearing request:
    • Require HTTPS.
    • Reject embedded credentials and malformed URLs.
    • Compare the normalized hostname and port against an explicit allowlist.
    • Resolve and reject loopback, link-local, private, and other unexpected destinations where custom public endpoints are not required.
  3. If self-hosted deployments must be supported, maintain an administrator-managed trusted-origin allowlist and require explicit confirmation before sending a token to a non-default origin.
  4. Prevent cross-origin credential forwarding during redirects. Reject redirects or verify every redirect target before resending authentication headers.
  5. Use narrowly scoped, short-lived, and revocable tokens so that disclosure has limited impact.
  6. Document clearly that changing ANYCAST_PORTAL_URL changes the party receiving ANYCAST_API_TOKEN.
  7. Avoid plain HTTP under all circumstances and enforce certificate verification.
  8. Add token rotation and revocation guidance for suspected endpoint misconfiguration or credential exposure.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: anycast-network
description: Connect to the Anycast agent network. List agents, query cross-environment connectors, send messages to remote agents, and check fleet status.
version: 0.1.0
metadata:
  openclaw:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README promotes capabilities like sending messages to remote agents and querying cross-environment connectors, including databases, APIs, and monitoring systems, but it does not warn users that these operations may access sensitive data or trigger impactful actions in other environments. In a skill specifically designed for cross-environment reach and remote agent interaction, the lack of safety guidance increases the risk of accidental misuse, overbroad access, or unsafe delegation by downstream agents and users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill performs authenticated network requests to a remote Anycast portal and can query cross-environment connectors, but it does not warn users that prompts, identifiers, and connector data may be transmitted off-host. In this context, the omission matters because the skill can access external systems and tenant data, so users may unknowingly trigger privacy- or impact-sensitive actions.

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 · SKILL.md (reported line 35)May include surrounding context.

Send a Message to an Agent

bash
curl -s -X POST \
  -H "Content-Type: application/json" \
  -H "x-agent-token: $ANYCAST_API_TOKEN" \
  "$ANYCAST_PORTAL_URL/api/agents/{agentId}/interrupt" \

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill exposes tenant-scoped memory set/get operations, but the manifest description does not disclose that it can persist and retrieve shared state. This can mislead users or higher-level safety systems about the skill's data handling capabilities, increasing the risk of unintended storage of sensitive or cross-session information.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill documents a memory write operation with scope set to tenant, but it does not warn that this can modify shared state visible across sessions, users, or agents depending on deployment. That omission can cause accidental persistence, poisoning of shared memory, or overwriting operational data without informed consent.

Content

No source excerpt is available for this finding.