Install
npx skills add squirrelscan/skillsWho this is for
This guide is for people who will hand website audits to an agent and accept that every skill is driven through the squirrel CLI. Officially it is written as website-audit skills, plugins, and MCP for agents, in one repository:
https://github.com/squirrelscan/skills
Current materials expand only two skills. squirrelscan operates the CLI: install, sign-in, run audits, publish reports, credits, API keys, MCP setup, configuration, and troubleshooting. Notes are at
https://github.com/squirrelscan/skills/blob/main/skills/squirrelscan/SKILL.md
audit-website is the fix loop: audit a site, map findings to code, change in batches, then audit again until the score improves. Notes are at
https://github.com/squirrelscan/skills/blob/main/skills/audit-website/SKILL.md
It fits when you can install the CLI and will only audit sites you are allowed to inspect. It does not fit if the CLI is not on PATH, or if you want the skills as CLI-free documents only.
Finish a first try in 3 minutes
Follow only the one official recommended path.
Precondition: this machine can run npx, and the squirrel CLI is installed and on PATH. The materials say every skill drives that CLI, but they do not give an install command for the CLI itself. If squirrel is not yet on PATH, let the squirrelscan skill handle install and sign-in first; do not jump to scanning.
Install location: handled by npx skills add. The materials do not write a skill-folder landing path; do not invent one that never appeared.
Copy and run:
npx skills add squirrelscan/skills
Success: the command finishes normally. First do a read-only check: ask the agent to see whether squirrel is on PATH and to describe the steps of one audit, but do not scan any URL, publish a report, or change code yet. If it can restate that the CLI is required and that sign-in / audit / publish are separate actions, the skill is loaded.
Before a real scan, confirm the target URL and that you are allowed to audit it. Scanning, publishing a report, and writing code all have side effects.
Usage cases
Case 1: Operate the CLI only; sign in first, then decide whether to audit. This matches squirrelscan. Fits when you have not run a scan yet and only want to confirm install, sign-in, configuration, and credits. Success: the CLI is usable and sign-in state is clear. The materials also mention hosted MCP:
https://mcp.squirrelscan.com/mcp
It uses streamable-http and can use OAuth or an Authorization: Bearer sq_... key; locally it is squirrel mcp. The first try need not attach MCP. Do not invent a complete key.
Case 2: After an audit, map to code, but stop at the mapping. This matches the audit-website fix loop. Fits when the CLI already runs and you want to see which files a confirmed site’s issues land in. Audit first and ask only for findings-to-code mapping; explicitly forbid batch edits. Success: you see issue-to-file pairing and the repo is still unchanged. After you confirm the patch scope, allow batch edits and a retest. The materials give no score threshold or batch size.
Usage limits
Current materials prove only the two skills above and one npx install command. Do not invent other skill names, and do not switch to an installer that never appeared. The materials do not name any required client.
Do not invoke these skills without the CLI. Publishing a report sends results off this machine; changing code writes the repo. Check first, then authorize. The key prefix in the materials is sq_...; do not invent an example key. For more, follow the official repository:
https://github.com/squirrelscan/skills