Scans skills and MCP servers for security threats using static analysis in Docker.
npx skills add https://github.com/joydai2026-del/skills --skill skill-scan-joydai2026-del
Please help me install this Agent Skill. Skill: skill-scan Source: https://github.com/joydai2026-del/skills/tree/main/skill-scan Command: npx skills add https://github.com/joydai2026-del/skills --skill skill-scan-joydai2026-del
--no-llm, so no API keys are needed and scanned content never leaves the machine. - Broad threat coverage: Checks roughly 19 threat categories including prompt injection, hidden instructions, data exfiltration, MCP tool-poisoning, supply-chain typosquats, and live CVE lookups. - Machine-readable gating: JSON output exposes risk_assessment.recommendation (SAFE / CAUTION / DO_NOT_INSTALL) and a 0-100 score for scripted install gates. - Use Case: Before installing a community skill from GitHub, point the scanner at the repo URL, get a risk score and top findings, and only proceed to a full manual review if the pre-screen passes. ## Quick Start Scan this skill folder with skill-scan and tell me the risk score, the top findings, and whether it is safe to install.High-intent search queries and answers about installing and using this skill.
Run SkillSpector in Docker against the skill folder: mount the target read-only and execute `docker run --rm -v "$TARGET:/scan:ro" skillspector scan /scan --no-llm`. You get a 0-100 risk score and a SAFE, CAUTION, or DO_NOT_INSTALL recommendation.
Pass the repository URL straight to the scanner: `docker run --rm skillspector scan https://github.com/user/some-skill --no-llm`. Nothing is written to disk, and the static analysis runs entirely inside the throwaway container.
The scan covers roughly 19 categories including prompt injection, hidden instructions, anti-refusal, data exfiltration, privilege escalation, memory poisoning, tool misuse, SSRF, supply-chain typosquats, and MCP-specific tool-poisoning and rug-pull checks, using static patterns, Python AST, taint analysis, YARA, and CVE lookups.
No. The mandatory `--no-llm` static mode needs no API keys and sends nothing off the machine except an anonymous package-name lookup to the public CVE database. The LLM mode is deliberately avoided because it can route scanned content to a provider.
No. A clean scan only passes the pre-screen. External skills should still go through a two-round review gate: a repository-security reviewer first, then a second different-vendor model, before installation.
nicepkgLocate Claude Code skills and MCP servers from marketplaces, GitHub, and registries.
AradotsoReview AI skill installations and external inputs for security threats.
SafeAgent-BeihangAnalyze a target Skill for security risks before installation.
sjc786526-coderOrchestrates subagents to review pull requests using code-review skills.