claude
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
claudecommand-line interface to execute tasks. It includes guidance on using flags such as--dangerously-skip-permissionsfor autonomous work, emphasizing that these should only be used in dedicated worktrees with mandatory manual diff reviews. - [INDIRECT_PROMPT_INJECTION]: The skill establishes a multi-agent workflow where the primary agent processes output from the
claudesub-agent. It incorporates specific safety protocols to mitigate risks from potentially untrusted or fabricated data: - Ingestion points: The primary agent ingests output from the
claude -pshell command execution as described inSKILL.md. - Boundary markers: The skill recommends structured output formats (
--output-format text/json) and explicit task contracts to delimit the sub-agent's work. - Capability inventory: The sub-agent has the capability to read files, grep, glob, and execute bash commands within the shared working tree.
- Sanitization: The instructions mandate manual verification of all sub-agent claims and diffs before they are accepted or relayed to the user.
Audit Metadata