LIFEHUBBER
Theme

AI Resources

SkillSpector

SkillSpector is NVIDIA's public scanner for AI agent skills, built for checking skill folders, files, repositories, URLs, and zip packages before people wire them into coding-agent workflows.

The README lists 68 vulnerability patterns across 17 categories, fast static analysis, optional LLM semantic evaluation, OSV.dev dependency lookups, terminal / JSON / Markdown / SARIF output, and Python 3.12+ setup through uv or pip. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.

What it is

Scanner for agent skills

SkillSpector is a command-line and LangGraph-based workflow for inspecting AI agent skill packages, including single SKILL.md files, local directories, zip files, URLs, and Git repositories.

Why it stands out

Agent-skill checks, not generic linting

The rule list is tuned to agent-skill problems such as prompt injection, data exfiltration, privilege escalation, tool misuse, memory poisoning, MCP least privilege, MCP tool poisoning, executable code, and dependency issues.

Availability

Repo, CLI, and report formats

The repository includes install instructions, scan commands, provider settings for optional LLM analysis, a development guide, tests, and report output paths for terminal, JSON, Markdown, and SARIF use.

Why it matters

What makes it useful

Reusable agent skills can carry instructions, tool behavior, dependencies, and risk patterns that are easy to miss by name alone. Its static checks, optional LLM review, OSV lookup, and report outputs give readers a concrete skill-inspection layer to compare.

Notable points

What stands out

The repository exposes the full inspection path: accepted inputs, static analyzers, optional LLM evaluation, OSV lookups, report formats, and known limitations. Its output supports human review; it is not a safety guarantee or a sandbox.

Before using

What to review

Which input style fits the workflow: local skill folder, single SKILL.md file, Git URL, regular URL, or zip file.

Whether to run static-only mode or configure the optional LLM analysis with OpenAI, Anthropic, NVIDIA build, or a local OpenAI-compatible endpoint.

How JSON, Markdown, or SARIF reports should be stored if they include snippets from private skill files or internal workflow code.

Which findings need human review before changing, rejecting, or publishing a skill.

Reader fit

Who may find it relevant

Builders trying agent skills across Claude Code, Codex CLI, Gemini CLI, or similar skills-aware tools.

Teams organizing reusable skills and wanting a report format that can fit local review, documentation, or CI workflows.

Less relevant for readers looking only for model releases, consumer chat apps, or general-purpose coding assistants.

Editorial note

Why LifeHubber lists it

LifeHubber lists SkillSpector because it turns agent-skill inspection into concrete artifacts: 68 patterns across 17 categories, optional semantic evaluation, dependency lookups, and terminal, JSON, Markdown, or SARIF reports that people can review or carry into CI.

Source links

Source materials

Reader note

Before relying on this entry

LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.

What to explore next

Move from finding a skill to deciding what it may do.

After scanning a skill, decide how resources are discovered, which later actions need runtime policy checks, and where a person must approve or stop computer control.

Related in LifeHubber

Keep the thread going

Follow the next layer with AI Resources for AI projects with original links and practical caveats, AI Guides for decision habits for messy AI choices, AI Access for free and low-cost ways to compare AI model access, AI Ballot for a clearer view of what readers are leaning toward, and AI Radar for AI stories that deserve a second look.