History & FAQ
Everything you need to know before you scan.
What is SkillTracer?
SkillTracer is a free scanner for AI-agent skills. When you submit a skill, it runs two kinds of analysis: static analysis that reasons about the skill's source rather than just pattern-matching the text, and dynamic analysis that detonates it to see what it actually does at runtime. You get a reproducible Threat score (benign, suspicious, or malicious) and Risk score (low, medium, high, or critical) that combine into a final recommended action, plus findings that quote their supporting evidence and map to both the OWASP Agentic Top 10 and MITRE ATLAS. Every scan is shareable as a public report.
Are the reports I submit public?
Yes. Every scan is published to a public feed at /reports. Anyone can read the report and download the source you submitted, with no account or login. There is no private-scan option today.
Treat running a scan as publishing. Do not submit secrets, API keys, credentials, or code you cannot share. To request removal of a scan, use the “Get in touch” button in the bottom-right corner.
What can I scan?
AI-agent skills. Submit a single SKILL.md (or any .md file) up to 500 KB, or a bundle of up to 50 files totaling 2 MB. You can upload files directly, or paste a GitHub link to either a single skill file or a whole repository.
Do you actually run my skill?
Both analyses run on every scan, inside a locked-down sandbox. Static analysis reads and reasons about your skill's source with no outbound network, and never executes the skill.
Dynamic analysis detonates the skill in an instrumented sandbox seeded with decoy tools and honeytokens, then observes what it actually does: what it accesses, what it calls, and what data it tries to send.
Is my code sent to anyone else?
SkillTracer sends your skill's source to Anthropic's Claude API, which performs the analysis. During static analysis the code is only read and reasoned about, never executed.
Do I need an account? Who can scan?
Reading public reports needs no account. Running a scan requires signing in with a work email; personal or free providers such as Gmail and Outlook are not accepted.
How many scans can I run per day?
Signed-in users get 25 scans per day, and quotas reset at midnight UTC. For a higher limit, contact us with the “Get in touch” button.
Which model should I choose?
Haiku is fastest, Sonnet is balanced (the default), and Opus is the deepest. When you upload a file, select two or more models to run a side-by-side comparison; each scans independently and counts as one scan against your quota.
What do the results mean?
Each scan rates a skill on two axes: a Threat score (benign, suspicious, or malicious) for how malicious the behavior is, and a Risk score (low, medium, high, or critical) for how severe the impact could be. Together they produce a final recommended action. Every finding quotes its supporting evidence and maps to both the OWASP Agentic Top 10 and MITRE ATLAS, with a coverage heatmap for each.
Can I download what I scanned?
Yes. The submitted skill can be downloaded from its report page, and because every report is public, anyone can download it. It saves as a file and is never rendered or executed.
How long do you keep reports? Can I delete one?
Reports are kept and remain public. An identical re-submission reuses a cached result for 30 days, after which it is scanned fresh. There is no self-service delete yet; to request removal, use the “Get in touch” button.
I found a false positive or a bug. What should I do?
Please tell us using the “Get in touch” button in the bottom-right corner of any page.
Is SkillTracer free? Is it open source?
SkillTracer is free to use. It is not open source.
Still stuck? Run a scan or browse public reports.