
SkillSpector
Reads an agent skill before you install it and reports what it would be able to do to your machine and your data
What is SkillSpector?
Installing a skill or an MCP server is closer to installing a browser extension than to adding a library: it arrives as instructions and scripts that an agent will follow with your credentials and your files in reach, and almost nobody reads them first. SkillSpector reads them. It scans a repository, archive or directory against a catalogue of patterns covering prompt injection, data being sent somewhere it should not go, privilege escalation, poisoned tool descriptions and supply-chain risk, then returns findings with a score and a recommendation. NVIDIA cites research finding vulnerabilities in 26.1% of skills examined and likely malicious intent in 5.2%. A pattern scanner reports suspicion rather than proof, so expect to review findings and record the ones you have accepted.
What can you do with SkillSpector?
- Answer one question: is this safe to install — Point it at a repository, a URL, an archive or a directory and it returns findings with a severity, a score out of a hundred and a recommendation.
- Look for the failures specific to agents — The pattern catalogue covers prompt injection, data exfiltration, excessive agency, memory poisoning and tool descriptions written to mislead — not generic code smells.
- Check the dependencies too — Known vulnerabilities in what a skill pulls in are looked up against a public advisory database, with an offline fallback when there is no network.
- Add a second, slower opinion — An optional stage sends the material to a model for a semantic read, which catches intent that a pattern match cannot express.
- Accept what you have already reviewed — Findings you have judged acceptable can be recorded in a baseline, so a re-scan shows only what is new instead of the same list again.
- Gate an install from a pipeline — Reports come out as JSON, Markdown or SARIF, so a build can fail on a finding and the result lands in the same place as other security checks.
Before you choose SkillSpector
- A pattern scanner reports suspicion, not proof. Expect findings on skills that are fine, which is why the tool ships a way to record the ones you have accepted — and a clean report is not a guarantee.
- The semantic stage that catches intent needs a model, and therefore a key and a bill; the fast static pass runs on its own but sees only what a pattern can describe.
Star history
27 Aug to 30 Oct · +8.5k
Frequently asked questions
Is SkillSpector free for commercial use?
SkillSpector is released under the Apache-2.0 licence — OSI-approved open source, which permits commercial use.
How can SkillSpector be deployed?
SkillSpector is available as Runs locally / Self-hosted.