orca-subagents
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow where subagents process the local workspace using the
--worktree currentflag and return results via temporary files in/tmp. This establishes a potential surface for indirect prompt injection if the subagents process malicious content from the repository, which is then passed back to the main agent context through the--report-pathmechanism. - Ingestion points: The main agent reads report files generated by subagents from
/tmp. Subagents ingest data from the local repository as part of their assigned tasks. - Boundary markers: The instructions do not define delimiters or safety instructions for the agent to distinguish between its own logic and potentially malicious content within these reports.
- Capability inventory: The skill interacts with multiple external CLI tools (
orca,claude,codex,grok,cursor-agent) and executes shell scripts. - Sanitization: No explicit sanitization or validation of the subagent's report content is mentioned before it is processed by the calling agent.
- [COMMAND_EXECUTION]: The
scripts/reachableshell script performs automated status checks by executing external CLI binaries includingclaude,codex,grok, andcursor-agent. These executions are limited to status and authentication checks of established tools and use safe quoting for the command names. - [EXTERNAL_DOWNLOADS]: The skill refers to the command
orca skills get orchestrationto fetch additional skill modules from the vendor's ecosystem. This is a standard mechanism for fetching external content within the tool's framework.
Audit Metadata