1. Find the source
If there is no repository, or the instructions are not public, you cannot check it. That is the case for 19,415 of the listings we indexed. Treat “no source” as a reason to slow down, not as neutral.
2. Read the instructions file
For a skill this is usually SKILL.md; for an MCP server, the README and the tool definitions. Look for instructions that tell the agent to hide actions, skip confirmation, or follow text from web pages or emails.
3. Search the code for five things
- Environment variables and key files (
process.env,os.environ,~/.ssh). - Network calls to hosts you do not recognise.
- Shell execution (
exec,spawn,subprocess). curl … | shstyle remote installs.- Sending email or messages without an approval step.
Each one can be legitimate. The question is whether it fits the job the skill claims to do.
4. Check permissions against purpose
A summariser that wants write access to your whole drive is a mismatch. In our scan, “asks for overly broad permissions” was one of the reasons for a Caution badge.
5. Pin the version
Install a specific commit, not “latest”. Botlens records the commit and a SHA-256 for every scanned listing so you can compare.
Or let the Botlens skill do the first pass for you.
FAQ
Is an MCP server different from a skill?
Yes. An MCP server is a running program that exposes tools; a skill is mostly instructions and optional scripts. Both run with your agent’s permissions, so the same checks apply.
What if I cannot read code?
Start with step 1 and step 4. No public source plus broad permissions is enough reason to skip.
Read next
- Is it safe to install AI agent skills?
- Prompt injection in agent skills, explained
- How to let your agent find tools via the Botlens API
Browse by category: Research · Coding & Dev · Design & Video · Writing & Content · Productivity · Finance & Crypto