manus
Pass
Audited by Gen Agent Trust Hub on Feb 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (LOW): The skill performs network operations to a non-whitelisted domain (
api.manus.ai). This is the intended primary purpose of the skill to facilitate communication with the Manus AI service. No sensitive local files are accessed or exfiltrated during these operations. - PROMPT_INJECTION (LOW): This skill is vulnerable to Indirect Prompt Injection (Category 8) due to its core delegation workflow.
- Ingestion points: The skill fetches text and file attachments via
GET /v1/tasks/{task_id}inSKILL.md. - Boundary markers: Absent. There are no delimiters or instructions to the host agent to ignore embedded commands in the Manus output.
- Capability inventory: The skill utilizes
curlfor network requests andjqfor data processing. - Sanitization: Absent. Output from the external agent is parsed and returned to the caller without filtering or escaping.
Audit Metadata