How do I install the Security Radar skill?

Run npx skillmds@latest add search-atlas-group/security-radar in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Security Radar safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Security Radar?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Security Radar free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Security Radar?

search-atlas-group (@search-atlas-group) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Security Radar by search-atlas-group

Name: Security Radar
Author: search-atlas-group

Security Radar

Audit your own agentic setup's security posture. Use when the user asks to check/harden their setup, run a security check on their machine, scan their installed Claude Code skills / MCP servers / project dependencies, or wants a posture brief. Cross-references the installed surface against OSV.dev + npm audit and the OWASP LLM/Web Top 10. Read-only.

File contents

security-radar

Continuous-monitoring counterpart to /security-scan. Where /security-scan vets a repo before you install it, security-radar watches the surface you already run — installed Claude Code skills, configured MCP servers, project dependencies, and permission configs — and cross-references them against live advisory feeds and the OWASP frameworks. It produces a plain-English posture brief.

It is read-only: it inspects config files and lockfiles and changes nothing.

When to use

"Run a security check on my setup" / "harden my setup" / "am I exposed?"
After installing new skills, MCP servers, or dependencies.
As a weekly habit (the threat surface changes even when your code doesn't — vetted packages get compromised later).

How to run

From the kit folder (or wherever you placed the script):

python3 security-radar.py            # audit deps in the current project + your global surface
python3 security-radar.py --path /path/to/a/client/project
python3 security-radar.py --json     # machine-readable output

No installs needed — it uses the Python standard library only. If a tool (npm) or the network (OSV.dev) is unavailable, that check is skipped and noted; the rest still runs.

What it checks

Installed skills (~/.claude/skills, ~/.claude/plugins) — static-scans each for risky shapes (remote-pipe-to-shell, eval of fetched code, hardcoded credentials). → OWASP LLM05 Supply Chain
MCP servers (~/.claude.json, Claude Desktop config) — flags 3rd-party / remote servers not on the trusted-vendor list that may mediate your auth tokens. → LLM07 Insecure Plugin Design
Permissions (~/.claude/settings.json) — flags bypassPermissions and unconstrained allow rules. → LLM08 Excessive Agency / A01 Broken Access Control
Dependencies (package.json / requirements.txt) — npm audit + OSV.dev cross-check for known-vulnerable versions. → LLM05 / A06 Vulnerable Components

How to present results

Run the script, then summarize the posture brief for the user: lead with the overall rating (CLEAN / CARE / QUARANTINE / REJECT), then walk the highest-severity findings first with the concrete fix for each. The full brief is also saved to ~/security-radar-report-<date>.md. Encourage a weekly cadence and designating one owner per team.

---name: security-radardescription: Audit your own agentic setup's security posture. Use when the user asks to check/harden their setup, run a security check on their machine, scan their installed Claude Code skills / MCP servers / project dependencies, or wants a posture brief. Cross-references the installed surface against OSV.dev + npm audit and the OWASP LLM/Web Top 10. Read-only.---# security-radarContinuous-monitoring counterpart to `/security-scan`. Where `/security-scan` vets a repo **before** you install it, `security-radar` watches the surface you **already** run — installed Claude Code skills, configured MCP servers, project dependencies, and permission configs — and cross-references them against live advisory feeds and the OWASP frameworks. It produces a plain-English posture brief.It is **read-only**: it inspects config files and lockfiles and changes nothing.## When to use- "Run a security check on my setup" / "harden my setup" / "am I exposed?"- After installing new skills, MCP servers, or dependencies.- As a weekly habit (the threat surface changes even when your code doesn't — vetted packages get compromised later).## How to runFrom the kit folder (or wherever you placed the script):```python3 security-radar.py # audit deps in the current project + your global surfacepython3 security-radar.py --path /path/to/a/client/projectpython3 security-radar.py --json # machine-readable output```No installs needed — it uses the Python standard library only. If a tool (`npm`) or the network (OSV.dev) is unavailable, that check is skipped and noted; the rest still runs.## What it checks1. **Installed skills** (`~/.claude/skills`, `~/.claude/plugins`) — static-scans each for risky shapes (remote-pipe-to-shell, eval of fetched code, hardcoded credentials). → OWASP LLM05 Supply Chain2. **MCP servers** (`~/.claude.json`, Claude Desktop config) — flags 3rd-party / remote servers not on the trusted-vendor list that may mediate your auth tokens. → LLM07 Insecure Plugin Design3. **Permissions** (`~/.claude/settings.json`) — flags `bypassPermissions` and unconstrained allow rules. → LLM08 Excessive Agency / A01 Broken Access Control4. **Dependencies** (`package.json` / `requirements.txt`) — `npm audit` + OSV.dev cross-check for known-vulnerable versions. → LLM05 / A06 Vulnerable Components## How to present resultsRun the script, then summarize the posture brief for the user: lead with the overall rating (CLEAN / CARE / QUARANTINE / REJECT), then walk the highest-severity findings first with the concrete fix for each. The full brief is also saved to `~/security-radar-report-<date>.md`. Encourage a weekly cadence and designating one owner per team.

Frequently asked questions