Manus Skill - Operate Manus through Openclaw

• Outcome: Pass
• Latest audit: May 29, 2026, 8:30 PM
• Version:
  1.0.2

Overview

This is a disclosed Manus API client, but it can spend credits, upload files, and store prompt metadata locally.

Install only if you trust this skill with your Manus API key. Add .manus_sessions.json to .gitignore, avoid putting secrets in prompts or uploaded files, and explicitly confirm new task creation, uploads, public sharing, connector use, webhooks, and any deletion or visibility-changing action.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Taint Tracking: Direct Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (8)

Tainted flow: 'upload_url' from requests.post (line 447, network input) → requests.put (network output)

Medium

• Category: Data Flow

Content

import mimetypes
        mime = mimetypes.guess_type(str(path))[0] or "application/octet-stream"
        with open(path, "rb") as f:
            requests.put(upload_url, headers={"Content-Type": mime}, data=f).raise_for_status()
        return record

    def list_files(self) -> dict:

• Confidence: 90% confidence
• Finding: requests.put(upload_url, headers={"Content-Type": mime}, data=f).raise_for_status()

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 90% confidence
• Finding: The skill clearly uses sensitive capabilities beyond simple documentation: it reads an API key from the environment, writes a persistent session file, and makes authenticated network requests to a third-party API, yet it declares no explicit permissions. That mismatch weakens reviewability and consent because operators may not realize the skill can transmit data externally and persist potentially sensitive prompts and task metadata locally.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The invocation text is extremely broad and includes catch-all triggers such as simple mentions of Manus, status checks, delegation phrasing, and references to IDs, making accidental activation more likely. In this skill, over-triggering is security-relevant because activation can lead to authenticated API calls, task creation that consumes credits, and possible transmission of user data to an external service.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The skill documents persistent storage of `.manus_sessions.json` containing prompts, task IDs, and URLs, and also supports file uploads and webhooks, but it does not provide a prominent upfront warning about privacy and external data handling. This is dangerous because users may unknowingly expose sensitive prompts, documents, or callback endpoints to local disk and a third-party platform.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The design explicitly stores `first_prompt` verbatim and recommends storing `last_prompt`, which can capture secrets, credentials, personal data, proprietary code, or other sensitive user inputs in a local plaintext registry. Because this is a session-tracking skill for an external agent platform, users are likely to send high-sensitivity content, so retaining raw prompts without an explicit privacy warning, minimization strategy, or protection controls materially increases the risk of local data exposure.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: This function transmits local file contents off-host with no user-facing warning, confirmation, or destination transparency beyond the code itself. In an agent skill that may be triggered on behalf of users, silent upload behavior increases the chance of accidental disclosure of sensitive local files.

Missing User Warnings

Medium

• Confidence: 86% confidence
• Finding: delete_task performs a destructive remote operation immediately with no confirmation, dry-run option, or guardrails. In a session-management skill where IDs may be reused across turns, accidental invocation could permanently remove remote work or metadata.

Missing User Warnings

Medium

• Confidence: 86% confidence
• Finding: delete_file deletes remote files without any visible confirmation or recovery protections. Because the skill manages uploaded artifacts, accidental or coerced deletion could cause data loss for users relying on stored files.

VirusTotal

51/51 vendors flagged this skill as clean.