orca-per-workspace-env
Fail
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill contains explicit instructions for the agent to bypass security boundaries. Specifically, it tells the agent that if a command fails with
runtime_access_denied, it should "re-run it with escalated permissions" to bypass sandbox blocks. - [COMMAND_EXECUTION]: The skill relies on the execution of various local binaries (
orca,orca-ide,orca-dev,orca-cli) to manage environments, check status, and retrieve metadata. It includes logic for the agent to resolve which executable to use based on environment variables likeORCA_CLI_COMMANDandORCA_DEV_REPO_ROOT. - [INDIRECT_PROMPT_INJECTION]: The skill utilizes a "discovery stub" pattern where the primary instructions are not contained in the skill itself but are fetched at runtime via the
ORCA skills getcommand. This creates a dependency where the agent's subsequent behavior is determined by the output of an external CLI tool, which could be exploited if that tool's output is compromised.
Recommendations
- AI detected serious security threats
Audit Metadata