ai-pro
Fail
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill provides explicit instructions and regex patterns for agents to search for sensitive credentials like 'SECRET', 'KEY', and 'PASSWORD' within a codebase. This facilitates the accidental discovery and exposure of authentication tokens in agent outputs. Evidence found in SKILL.md ('git grep -E "(SECRET|KEY|PASSWORD)"') and references/architect-archive.md ('git grep -E "AUTH_.*_SECRET"').
- [INDIRECT_PROMPT_INJECTION]: The skill implements an autonomous agent architecture that ingests external codebase archives (via Repomix) and search results. While documentation suggests safety gates, the provided implementation lacks robust sanitization of these ingestion points, leaving the agent vulnerable to instructions embedded in the processed data. Evidence in references/ai-agents.md and references/architect-archive.md.
- [COMMAND_EXECUTION]: The skill makes use of various system commands for repository management, secret generation, and codebase analysis. Evidence includes usage of 'git grep', 'openssl', and 'bun x'.
- [EXTERNAL_DOWNLOADS]: The skill relies on and recommends downloading external packages and libraries. Evidence includes the use of 'bun x repomix' for context packing and instructions to install the 'requests' Python library.
Recommendations
- AI detected serious security threats
Audit Metadata