orca-cli
Fail
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill contains explicit instructions for the agent to bypass sandbox restrictions. Specifically, it tells the agent that if a command is blocked by the sandbox (resulting in a
runtime_access_deniederror), it should re-run the command with escalated permissions. - [DYNAMIC_EXECUTION]: The skill uses a pattern where the core logic and instructions are not present in the static file but are instead fetched at runtime using the command
ORCA skills get orca-cli. This dynamic instruction loading makes the full behavior of the skill opaque to static analysis and can be used to inject arbitrary instructions after the skill is loaded. - [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection by processing untrusted data from multiple sources.
- Ingestion points: The skill reads data from Orca-managed terminals, worktree comments, and an embedded browser (SKILL.md).
- Boundary markers: There are no instructions or delimiters provided to help the agent distinguish between its core instructions and embedded malicious content in these data streams.
- Capability inventory: The agent can execute commands via the
orcaCLI, manage filesystems/worktrees, and interact with web pages. - Sanitization: No sanitization or validation of the external content is mentioned before it is processed by the agent.
- [COMMAND_EXECUTION]: The skill executes local binaries (
orca,orca-ide,orca-dev) based on environment variables such asORCA_CLI_COMMANDandORCA_DEV_REPO_ROOT, allowing the environment to dictate which code is executed.
Recommendations
- AI detected serious security threats
Audit Metadata