computer-use
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from untrusted external sources, specifically the user interface of third-party desktop applications.
- Ingestion points: According to
SKILL.md, the agent reads accessibility trees and takes screenshots of local apps, including communication tools like Slack and web browsers. - Boundary markers: The skill instructions do not define clear delimiters or instruct the agent to ignore potentially malicious commands embedded within the UI content it processes.
- Capability inventory: The agent is granted extensive control capabilities, such as clicking, typing, dragging, and setting values, which could be exploited if it follows instructions found within the applications.
- Sanitization: There is no evidence of filtering or sanitization of the text extracted from the applications before it is processed by the agent's logic.
- [COMMAND_EXECUTION]: The skill executes local command-line tools (
orca,orca-dev,orca-ide) to perform desktop actions. It uses logic involving environment variables likeORCA_CLI_COMMANDandORCA_DEV_REPO_ROOTto determine which binary to run. - [DATA_EXFILTRATION]: The skill accesses sensitive user data by reading the state and visible UI of desktop applications like Slack and Spotify. While no network exfiltration is explicitly performed by the skill's instructions, the exposure of this data to the agent's context is a prerequisite for the skill's core functionality.
Audit Metadata