What does this skill do, and when should you use it?
The squirrelscan skill lets an AI agent operate the squirrelscan website audit tool end to end: installing the CLI, authenticating, running audits, publishing reports, managing cloud credits, configuration, and troubleshooting. The tool crawls a site like a search engine, analyzes every page against 260+ rules in 21 categories, and returns a health score with concrete, fixable issues. Local audits are free and run entirely on your machine with no account needed; cloud features such as report publishing, browser rendering, and scheduled crawls are pay-as-you-go with credits.
Runs squirrel audit <url> to crawl, analyze, and report in one step, with quick (25 pages), surface (100), and full (500) coverage modes. Produces output in console, text, , html, markdown, xml, and a token-optimized --format llm for agents. Generates regression diffs between audits (report --diff), publishes shareable reports to reports.squirrelscan.com, and supports CI gating via --fail-on "score<90". Also connects agents over MCP via squirrel mcp (local stdio) or the hosted server, and exposes the send_feedback tool for mid-session bug reports.
- A developer wants a pre/post-deploy health check and asks the agent to run `squirrel audit https://example.com --format llm` and summarize critical issues
- A CI engineer wants to block quality regressions, configuring SQUIRRELSCAN_API_KEY and `--fail-on "score<90"` in a headless pipeline
- A team investigating exposed secrets (96 patterns including OpenAI, AWS, Stripe) scans their site with the security rules
- A maintainer low on disk space reclaims it with `squirrel self disk --prune --keep 3`
- An agent on an MCP-capable client queries audit results live via the hosted server at mcp.squirrelscan.com
- A platform behind Shopify/Cloudflare bot protection authorizes crawls using custom Web Bot Auth headers
How do you install this skill?
- Core cloud features (report publishing, browser rendering, hosted MCP) rely on overseas endpoints with no stated mainland-China reachability; use --offline when isolation matters.
- Audits publish to the cloud (unlisted) by default and include telemetry; use --no-publish/--offline if data egress is a concern.
- Self-update downloads a binary from an external source; verify provenance in a controlled environment before running.
- Publisher identity is unverified; key capability claims (260+ rules, 96 secret patterns) are author assertions, not independently validated.
- Pruning history (--prune) irreversibly retires old audits; agents should follow the doc's requirement to confirm with the user first.
- Shell / CLI
- Network access
- Local filesystem
- MCP Server
squirrel CLI
Prerequisite: install the squirrel CLI (squirrelscan.com/download); the binary lands in ~/.local/bin/squirrel and must be in PATH. Install the skill via:
- Any agent:
npx skills add squirrelscan/skills(add--skill squirrelscanfor just this one) - Claude Code:
/plugin marketplace add squirrelscan/skillsthen/plugin install squirrelscan@squirrelscan - squirrel CLI:
squirrel skills install - Manual: clone the repo and symlink/copy skills/squirrelscan into your agent's skills directory.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/squirrelscan/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/squirrelscan" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Audit example.com and show me the top 5 critical issues
- Check my site for broken links and leaked secrets.
Example prompts: "Audit example.com and show me the top 5 critical issues" or "Check my site for broken links and leaked secrets." Common commands:
bash
squirrel init -n my-project
squirrel audit https://example.com --format llm # quick mode (25 pages) by default
squirrel audit https://example.com -C full -m 500 --format llm
squirrel report --list
squirrel report <audit-id> --format markdown -o report.md
squirrel auth login # needed for cloud features; CI uses SQUIRRELSCAN_API_KEY
Local audits are free; add --offline for no cloud contact at all. Troubleshooting: on 401, re-run squirrel auth login or check the API key; if the command is not found, ensure ~/.local/bin is in PATH.
What are this skill's strengths and limitations?
- Local audits are free, run fully on your machine, need no account, and support a fully offline --offline mode
- LLM-native output (--format llm) is token-optimized for agent consumption
- 260+ rules across 21 categories, including practical secret-leak detection
- Both local stdio and hosted MCP integration, plus solid CI gating and regression diffs
- Audits are cached locally, so reports re-render without re-crawling
- Requires the squirrel CLI installed and in PATH; the skill cannot function without it
- Cloud features (browser rendering, publishing, scheduled crawls) cost credits; exact pricing requires running `squirrel credits`
- Full mode caps at 500 pages by default, limiting very large site coverage
- Offline mode excludes cloud rendering and report publishing
- No independent test suite or third-party evaluation evidence in the source material
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Squirrelscan Website Audit Skill this page | 53 · Use with care | ★ 99 | 12d ago | MIT |
| squirrelscan Website Audit & Fix Loop | 49 · Use with care | ★ 99 | 12d ago | MIT |
| Pulser — SKILL.md Linter | 50 · Use with care | ★ 18 | 3mo ago | MIT |
| Agent Reach | 62 · Recommended | ★ 97k | 18d ago | MIT |
| skill-optimizer: Eval Workbench for Agent Skills | 61 · Recommended | ★ 81 | 5mo ago | MIT |
How did FollowSkills review this skill?
Least-privilege allowlist (Bash(squirrel:*) and Read), clear secret-handling guidance (never commit, use secret store, output redaction), and escape hatches (--offline/--no-publish), confirm-before-prune with --dry-run; data flows (cloud publish, telemetry, credit billing) largely disclosed. Deducted for: default publishing to cloud (unlisted) and default telemetry/billing paths, self-update pulling a binary from an external source, unverified publisher identity, and a broad squirrel:* scope that includes keys/auth.
Docs are self-consistent with a command table, troubleshooting (command not found, 401, slow crawl, invalid URL), and self doctor feedback paths; failure messages are well designed. Deducted for: no committed tests or reproducible validation of key paths; abnormal-input behavior is mostly prose; static review cannot execute, capped at 10.
Trigger description is clear (when to use, boundary vs companion audit-website skill, --format llm for agents); inputs, output formats, coverage modes and non-fit ranges are mostly clear. Deducted for: cloud features (publishing, rendering, hosted MCP) depend entirely on overseas endpoints with no mainland-China reachability statement; version-dependent behavior (requires 0.0.92+) is a hidden prerequisite.
Well-layered structure (install → commands → quickstart → auth → reports → maintenance → troubleshooting), MIT license, version 1.3, explicit maintenance commands and update path. Deducted for: no changelog, no centralized known-limitations disclosure, thin version governance, and author attribution is a name without verifiable ownership.
Goal is clear (audit a site, return fixable issues); llm output format, diffs, and CI fail-on thresholds are directly agent-usable, with a free local mode. Deducted for: no representative output samples verifiable in static review, core value depends on installing an external binary and unverified cloud services, and comparative-benefit evidence is thin.
Key claims (260+ rules, 96 secret-detection patterns) are author statements only; no third-party execution evidence, committed tests, or CI in the provided files. Only internal consistency of repo, license, and docs serves as weak corroboration, hence below 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →