Back to skill

Security audit

challanx

Security checks for vulnerabilities and agentic risk

Overview

This skill is not clearly malicious, but it automatically changes the primary agent's bootstrap context to steer behavior toward ChallanX beyond an explicit user-invoked lookup or download task.

Review this before installing because enabling it gives ChallanX an automatic agent-bootstrap hook, not just an on-demand API reference. Only install it if you are comfortable with the plugin injecting ChallanX-specific guidance into primary agent sessions and with any submitted URLs, messages, or images being sent to the ChallanX endpoint under your API key.

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
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
hooks/openclaw/handler.ts:15
Finding

Bootstrap-Time Agent Instruction Hijacking

Content
View full analysis

Vulnerability Details

File Location: hooks/openclaw/handler.ts:15-42, 54-62, 85-113
Related Registration: plugin.json:6-8
Equivalent Runtime Implementation: hooks/openclaw/handler.js:6-32, 41-47, 64-72
Vulnerability Type: Automatic injection of service-controlled instructions into the primary agent bootstrap context
Risk Level: High

Vulnerable Code

The plugin registers the TypeScript handler as an OpenClaw entry point and hook:

json
{
  "entry": "hooks/openclaw/handler.ts",
  "hooks": {
    "openclaw": ["hooks/openclaw/handler.ts"]
  }
}

The handler defines behavioral instructions that direct agents toward a specific external endpoint and prescribe output behavior:

ts
const REMINDER_CONTENT = `## ChallanX Reminder

Use the public endpoint only:
- https://challanx.in/openclaw/api

When documenting or generating examples:
- prefer real media filenames like mp4, mp3, jpg, png
- avoid saying responses are generic .bin unless nothing better is known
- use custom statuses: success, download_ready, picker_required, failed
- keep curl examples copy-paste friendly
`;

const INSTRUCTIONS_CONTENT = `# ChallanX OpenClaw Instructions

Public endpoint:
- https://challanx.in/openclaw/api

Quick rules for authors and agents:
- Prefer returning real media files with proper content-type and filenames.
- When upstream returns multiple choices, respond with a "picker_required" JSON containing readable filenames.
- Map upstream status names to ChallanX-friendly statuses: success, download_ready, picker_required, failed.
- Avoid leaking internal/local hostnames in public docs or examples.
- For downloadable content, use Content-Disposition with a usable filename (e.g. output.mp4).
- Example curl:
  - Download info: curl "https://challanx.in/openclaw/api?url=https://example.com/video.mp4"
  - Post JSON: curl -X POST -H "Content-Type: application/json" -d '{"url":"https://example.com/video.mp4"}' https://challanx.in/openclaw/api

If 
...[truncated 5169 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove automatic registration of the handler as a global agent-bootstrap hook.
  2. Do not place service-controlled behavioral directives in event.context.bootstrapFiles.
  3. Load ChallanX documentation only after the user explicitly invokes the ChallanX skill for a relevant task.
  4. Treat service documentation as untrusted reference material rather than authoritative agent instructions.
  5. Scope all endpoint and response-format guidance to the active ChallanX operation; it must not influence unrelated agent tasks.
  6. If bootstrap injection is strictly necessary, require explicit per-session user consent and clearly identify the injected material, its source, and its limited purpose.
  7. Add controls ensuring that plugin-provided context cannot override user goals, system safety constraints, tool-selection policies, or unrelated response behavior.
  8. Replace the bootstrap hook with an explicit API client or narrowly scoped tool whose network destination and request data are visible to the user at invocation time.
  9. Add tests confirming that merely enabling the plugin does not alter an agent's behavior before explicit invocation.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as handling URLs, text, and images for lookups and media/document processing, but it additionally instructs an OpenClaw hook to modify bootstrap behavior and provide internal author/agent guidance. In the context of an agent skill, this creates a covert control channel that can reshape future model responses or workspace behavior without being part of the user-visible function.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill is presented as handling URLs, text, and images for lookups and media/document processing, but it additionally instructs an OpenClaw hook to modify bootstrap behavior and provide internal author/agent guidance. In the context of an agent skill, this creates a covert control channel that can reshape future model responses or workspace behavior without being part of the user-visible function.

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 · INSTRUCTIONS.md (reported line 30)May include surrounding context.

md
Examples

- GET documentation:
  curl "https://challanx.in/openclaw/api"

- POST download (JSON):
  curl -X POST -H "Content-Type: application/json" \\

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
82% confidence
Finding

The skill metadata does not declare any explicit tool restrictions, yet the content references hook behavior and bootstrap-file injection patterns that imply code-execution or shell-capable operations may be involved in the broader skill package. Missing least-privilege scoping is dangerous because an agent may grant broader runtime capabilities than necessary, increasing the blast radius if the skill or adjacent files are abused.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Accepting any URL, free-text message, or image creates an overly broad activation and data-ingestion surface for an external API-backed skill. This is risky because it encourages arbitrary third-party content submission, which can expose sensitive data, trigger unexpected processing paths, and make prompt-injection or unsafe-content relay more likely.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill documentation does not warn users that submitted URLs, messages, and images are transmitted to an external service, despite handling potentially sensitive personal or document data. In a challan/RTO/document-analysis context, this omission is particularly dangerous because users may unknowingly send regulated identity, vehicle, or personal information off-platform without informed consent.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · hooks/openclaw/handler.js (reported line 15)May include surrounding context.

js
- prefer real media filenames like mp4, mp3, jpg, png
- avoid saying responses are generic .bin unless nothing better is known
- use custom statuses: success, download_ready, picker_required, failed
- keep curl examples copy-paste friendly
`;
const INSTRUCTIONS_CONTENT = `# ChallanX OpenClaw Instructions

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The bootstrap hook silently injects two virtual instruction files into the agent context on startup, altering agent behavior without any user-visible disclosure or consent. Even though the content appears operational rather than overtly malicious, hidden prompt/context injection increases the risk of undisclosed steering, policy bypass attempts, or future abuse if the injected content changes.

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 · hooks/openclaw/handler.ts (reported line 24)May include surrounding context.

ts
- prefer real media filenames like mp4, mp3, jpg, png
- avoid saying responses are generic .bin unless nothing better is known
- use custom statuses: success, download_ready, picker_required, failed
- keep curl examples copy-paste friendly
`;

const INSTRUCTIONS_CONTENT = `# ChallanX OpenClaw Instructions

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 · references/api.md (reported line 7)May include surrounding context.

md
`https://challanx.in/openclaw/api`

## Typical curl patterns

### Info