orca-per-workspace-env
Fail
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill explicitly directs the agent to attempt to bypass access restrictions by using escalated permissions. * Evidence (SKILL.md): 'If a command fails with
runtime_access_denied, your sandbox blocked the connection: re-run it with escalated permissions.' - [COMMAND_EXECUTION]: The skill instructs the agent to identify and execute various local binaries (
orca,orca-dev,orca-ide) which have the capability to create and manage VMs, SSH hosts, and containers. * Evidence (SKILL.md): Logic provided for resolving theORCAexecutable from environment variables or specific paths. - [INDIRECT_PROMPT_INJECTION]: The skill acts as a stub that ingests and obeys instructions generated at runtime by a local command-line tool. * Ingestion points: The content returned by
ORCA skills get orca-per-workspace-env(SKILL.md). * Boundary markers: Absent; there are no delimiters or instructions provided to the agent to treat the tool's output as untrusted data. * Capability inventory: Access to the local shell and the ability to execute environment management commands (VM/Container creation). * Sanitization: Absent; the agent is instructed to 'load the version-matched guide' and follow it. - [DYNAMIC_EXECUTION]: The skill uses a dynamic loading pattern where the actual instructions followed by the agent are fetched during the session rather than being statically defined in the skill file. * Evidence (SKILL.md): 'This discovery stub loads the version-matched guide from the Orca executable used for this session.'
Recommendations
- AI detected serious security threats
Audit Metadata