PromptArmor Claude Skills Risk Assessment

Claude Skills Risk Assessment is a threat intelligence service by PromptArmor that evaluates the security and privacy risks of Claude skill definitions — the configuration files and tool descriptions that define how AI agents behave and what they can access. Users submit a skill definition via GitHub repo URL, Claude directory link, raw SKILL.md URL, file upload, or direct paste. PromptArmor's threat intelligence team then assesses the skill across multiple risk dimensions. Assessment framework covers the following categories: - Write Actions: Identifies skills that can take irreversible actions (file writes, email sends, API POST requests, code execution, calendar mutations) that could be triggered at machine speed via a single compromised prompt. - Skill Definition Injection: Detects adversarial instructions embedded directly inside the skill definition itself — within the system prompt, tool descriptions, or SKILL.md — which execute with full agent trust at load time. - Authentication Scope: Evaluates whether a skill inherits ambient authentication from already-authenticated applications (Gmail, Slack, GitHub, cloud storage) without initiating its own OAuth flow, thereby silently assuming the user's full identity scope. - Data Exfiltration: Identifies skills that combine read access to sensitive data (emails, documents, databases) with external endpoint access, creating exfiltration chains that bypass traditional DLP controls. - Tool Chaining: Assesses multi-step agentic workflows where a compromised tool call can cascade into downstream write actions, with blast radius growing at each tool-to-tool trust boundary. - Prompt Injection Surface: Flags skills that fetch or process external content (web pages, documents, GitHub issues, emails), which can expose an injection surface for adversarial instructions to hijack agent behavior mid-task. - Privacy & PII Exposure: Evaluates access to calendars, contacts, HR systems, or internal documents for potential PII over-exposure and violations of GDPR or HIPAA data minimization requirements. The tool emphasizes that risk assessments must evaluate combinations of capabilities rather than individual tools in isolation.