orca-emulator-android
Fail
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructions explicitly direct the agent to re-run commands with escalated permissions (e.g., sudo or administrative rights) if a 'runtime_access_denied' error is encountered. This encourages the agent to bypass security boundaries and provides a path for elevated execution of potentially untrusted or unverified tools.
- [COMMAND_EXECUTION]: The skill relies on executing a local CLI tool resolved through a hierarchy of environment variables (ORCA_CLI_COMMAND, ORCA_DEV_REPO_ROOT) and local paths (orca-ide, orca). This mechanism can be exploited if an attacker manipulates the environment variables to point to a malicious binary.
- [INDIRECT_PROMPT_INJECTION]: The skill serves as a 'discovery stub' that loads a 'version-matched guide' using the 'ORCA skills get' command. This creates a vulnerability where the content returned by the external tool, which the agent is instructed to follow, is not sanitized and lacks boundary markers to prevent the execution of embedded instructions.
- Ingestion points: The output of the command 'ORCA skills get orca-emulator-android' is ingested to provide the operational logic for the session (SKILL.md).
- Boundary markers: There are no delimiters or warnings to ignore embedded instructions within the content fetched from the external CLI tool.
- Capability inventory: The skill description indicates broad capabilities, including ADB control, application installation, logcat access, accessibility tree interaction, and hardware button simulation.
- Sanitization: No sanitization, validation, or filtering is performed on the content retrieved from the external source before the agent processes it.
- [DYNAMIC_EXECUTION]: The skill uses a local executable to dynamically retrieve and load the actual instructional guide at runtime, rather than providing the instructions statically within the skill files.
- [EXTERNAL_DOWNLOADS]: The 'skills get' command pattern represents a staged loading process where external content is fetched and executed/followed by the agent, potentially from unverified remote sources managed by the CLI tool.
Recommendations
- AI detected serious security threats
Audit Metadata