skill-injection-defense-adrian-lerer

lawvable/awesome-legal-skills

Audits legal AI skills, prompts, workflows, MCP/tool instructions, and agent packages before they are trusted or installed. It detects prompt injection, hidden or hostile instructions, unsafe scripts, suspicious frontmatter, credential exposure, exfiltration paths, persistence mechanisms, cron/launchd hooks, unauthorized network calls, and supply-chain risk. Use it before adopting third-party or generated skills; before publishing a skill to a legal AI marketplace; when reviewing SKILL.md files, skill folders, scripts, references, MCP manifests, or automation instructions; or when a user asks whether an AI workflow is safe. Paste or upload the skill, prompt, MCP/tool instruction, etc you want to review. The auditor treats the material as untrusted data: it does not execute scripts, install packages, or obey commands embedded inside the reviewed content. It returns a verdict — approve, approve with constraints, rewrite, quarantine, or reject — plus the key risks and recommended remediation.

GitHub stars

900

Sponsors

SnugThe clipboard manager you and your agent can use, safelyAgent CaffeineClose your lid, come back to finished workWaxBlock anything with supervision, without erasing your iPhoneOpen StepsSponsoredFree and open-source skills that make AI coding agents easier to understand, verify, and control.