jikkyo
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes the
gh(GitHub) andgitCLI tools to manage repository worktrees and issue comments. These operations are safely constrained by absolute path script invocations and strict regex-based validation of the GitHub issue URL, preventing command injection. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted data from GitHub issue comments and sub-agent output. This attack surface is mitigated by comprehensive logic:
- Ingestion points: External data enters via
gh apicalls for issue comments and the standard output stream of the headless sub-agent. - Boundary markers: The system uses machine-readable hidden HTML comment markers (
<!-- jikkyo:status ... -->) to distinguish its own status updates from user-generated content. - Capability inventory: The skill's capabilities are limited to specific GitHub API interactions (listing and posting comments) and local git worktree management.
- Sanitization:
SKILL.mdincludes explicit instructions for the agent to treat issue comments and external messages as data rather than instructions, and to quote back suspicious text instead of acting on it. - [REMOTE_CODE_EXECUTION]: The skill launches a headless
claudeinstance to perform tasks. This is a core intended feature and is secured using aDENY_RULESlist that explicitly blocks sensitive operations like force pushes, branch deletions, or API-based comment manipulation by the child process.
Audit Metadata