What problem does it solve? Installing a third-party agent skill grants it Read, Bash, and network access under your identity, yet most users install skills without any security review. This Skill gates that decision by combining a deterministic CLI scan with structured manual review checklists so malicious or leaky skills are caught before they run. ## Core Features & Use Cases - Deterministic first pass: Runs rafter skill review against a local path or git URL to extract secrets, URLs, high-risk shell patterns, obfuscation signals, and frontmatter tool grants into a JSON report. - Branch-specific review playbooks: Provides separate workflows for new installs, updates to trusted skills, and investigations of suspicious skills, each pointing to focused sub-docs on provenance, malware indicators, prompt injection, data practices, telemetry, and changelog diffs. - Use Case: Before installing a community SKILL.md from GitHub, run the review command, then walk the authorship-provenance and prompt-injection checklists to confirm the author is legitimate, no hidden instructions exist, and allowed-tools matches the stated purpose. ## Quick Start Ask the agent to vet a third-party skill before installing it, for example: review the skill at ./third-party-skill/ and tell me whether it is safe to install.
