coding-agents-superpack
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for running discovery commands (
which,--version) and health checks (doctor) for various coding agents like Claude Code, Codex, and Grok CLI. These are standard diagnostic operations for environment discovery. It also references automation flags such as--always-approvefor specific CLI tools, which are documented as examples for headless execution patterns.- [INDIRECT_PROMPT_INJECTION]: The skill architecture relies on agents reading from and writing to shared artifacts likeKANBAN_BOARD.mdandSTATE.jsonto facilitate collaboration. This creates a potential surface for indirect prompt injection if external or untrusted data is processed through these shared files.\n - Ingestion points: Shared workspace files (
KANBAN_BOARD.md,STATE.json) used for multi-agent state management.\n - Boundary markers: The skill suggests structured handoff artifacts but does not explicitly define prompt isolation or injection delimiters for external content.\n
- Capability inventory: The skill is designed to drive CLIs with tool access, including file reading, editing, and shell execution.\n
- Sanitization: Explicit instructions are provided to avoid placing secrets in prompts or shared files, reducing the risk of credential exposure during multi-agent interactions.
Audit Metadata