Ralph Ultra Security Audit

Deep-dive security audit with 1,000 iterations (~4-8 hours). Use when user says 'deep security audit', 'ralph ultra', 'compliance audit prep', 'thorough secu...

This skill is a powerful, open-ended audit runner: it tells the agent to read code, environment variables, git history, database constraints, and to generate proof-of-concept exploits. That is coherent for a deep audit, but the skill metadata does not declare what credentials or files it will need — so installing it could allow the agent to access sensitive data without you knowing upfront. Before installing or running: (1) only run in an isolated/staging environment or sandbox with no production secrets; (2) require the author to document exact inputs, expected repo paths, and credential requirements; (3) restrict the agent's filesystem/DB access to read-only snapshots where possible; (4) decide whether you permit automatic PoC generation (it can create exploit code); and (5) consider manual human review of findings before any remediation is applied. If you need higher assurance, ask the publisher to add explicit required env/config entries and scope constraints in the SKILL.md so required privileges are visible before installation.

Type: OpenClaw Skill Name: ralph-ultra Version: 3.0.0 The skill bundle is classified as suspicious due to the extensive and high-risk capabilities it instructs the AI agent to perform, even though these actions are framed within the context of a 'deep security audit'. Instructions in `SKILL.md` include explicit command execution (`git rev-parse --show-toplevel`), broad file system access (reading code, `.env` files, git history for 'secret detection', CI/CD configs, and writing to `.ralph-report.md`), network enumeration ('endpoint enumeration', 'exposed ports'), and 'penetration test simulation' with 'Red Team Mindset' and 'proof-of-concept' generation. While these actions are necessary for a thorough security audit, they represent significant potential for abuse or unintended harm if the agent were compromised or if the instructions were subtly subverted, making it a high-risk tool rather than benign.