founder-mode
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to use command-line interfaces such as 'claude', 'codex', and 'npx' to perform installation, updates, and removal of plugins. These operations are specifically scoped to the management of the vendor's 'guava' product suite and the 'skills' utility.
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external platforms including Notion, Google Drive, and local Markdown folders where the 'founder-mode' rules are stored. This data ingestion path constitutes an indirect prompt injection surface.
- Ingestion points: External storage locations used for the 'guava-hub' (Notion, Google Drive, OneDrive, Obsidian).
- Boundary markers: Instructions specify placing rules within a dedicated 'How to work with me' section to distinguish them from other content.
- Capability inventory: The skill modifies the agent's core operating principles and executes shell commands for plugin maintenance.
- Sanitization: Rule 26 explicitly directs the agent to check third-party skills against 'skills.sh' and perform security reviews before installation, mitigating risks from untrusted external content.
Audit Metadata