Audits a Claude Code skill for security risks in three modes: before download (from a URL or install command), after download but before install (from a .skill file), or after install (from a local skills directory). Use this skill whenever a user is about to install a skill from any source — including GitHub URLs, git clone commands, npx/npm commands, curl/wget downloads, pip installs, marketplace links, or raw SKILL.md URLs. Also trigger when a user asks "is this skill safe?", "should I trust this skill?", "can you check this before I install it?", "audit this skill", or pastes any link to a skill repository or .skill file. If a user mentions installing ANY skill, proactively offer to audit it first — do not wait for them to ask.
97
97%
Does it follow best practices?
Impact
99%
1.28xAverage score across 5 eval scenarios
Advisory
Suggest reviewing before use
{
"context": "Tests whether the agent correctly identifies WARNING-level findings in a Mode 2 (downloaded, not installed) audit: Bash access (A1), overly broad tool list (A4), script with file access outside working directory (B5), and outbound network calls (B2 standalone). Verdict should be PROCEED WITH CAUTION. Mode 2 transparency notice should appear.",
"type": "weighted_checklist",
"checklist": [
{
"name": "PROCEED WITH CAUTION verdict",
"description": "The report states a PROCEED WITH CAUTION verdict (or equivalent — warnings found, no criticals)",
"max_score": 10
},
{
"name": "Mode 2 transparency notice",
"description": "The report contains a transparency notice matching Mode 2: states the .skill file or extracted file was read, treated as data only, and advises verifying the file source is trusted",
"max_score": 10
},
{
"name": "A1 Bash access flagged",
"description": "The report flags Bash in the allowed-tools list as a warning (A1 or equivalent)",
"max_score": 10
},
{
"name": "A4 broad tool list flagged",
"description": "The report flags the broad tool list (Read Write Bash WebFetch Glob Edit = 6 tools) as a warning (A4 or equivalent)",
"max_score": 10
},
{
"name": "B5 out-of-directory access flagged",
"description": "The report flags file access outside the working directory, specifically the script's reference to ~/.config/report-tool/config.json",
"max_score": 12
},
{
"name": "Script reviewed",
"description": "The report explicitly states that scripts/format_report.sh was read and reviewed",
"max_score": 8
},
{
"name": "Remediation steps included",
"description": "The report includes at least one concrete remediation step for at least one warning (not just identifying the issue)",
"max_score": 10
},
{
"name": "Plain-language framing",
"description": "The report uses plain-language explanations appropriate for a non-technical reader (the product manager), not just technical check codes",
"max_score": 10
},
{
"name": "What Was Reviewed section",
"description": "The report lists both SKILL.md and the script (format_report.sh) in the reviewed section",
"max_score": 10
},
{
"name": "Static audit reminder",
"description": "The report includes the standard reminder that a clean or cautionary audit is not a guarantee of safety",
"max_score": 10
}
]
}audit-sample
evals
scenario-1
scenario-2
scenario-3
scenario-4
scenario-5