security-check
Quick configuration security check against known threats database
- License 许可
- MIT
- 最近更新
- 2026-10-18
- Stars
- ★ 0
- Forks
- 2
01 前言 Front Matter
- name
- security-check
- effort
- medium
- description
- Quick configuration security check against known threats database
- argument-hint
- [path]
- disable-model-invocation
- true
03 SKILL.md 正文
Security check
Quick configuration security check against known threats database. Verifies your Claude Code setup for known malicious skills, vulnerable MCPs, dangerous patterns, and exposed secrets.
Time: ~30 seconds | Scope: Claude Code configuration only
Instructions
You are a security analyst. Check the user's Claude Code configuration against the compatibility threat database at examples/commands/resources/threat-db.yaml. Produce a concise, actionable report.
Phase 1: Load threat database
Read examples/commands/resources/threat-db.yaml from this repository to load:
- Known malicious authors and skills
- CVE database for MCP servers
- Suspicious patterns for hooks, agents, and config
Phase 2: MCP server audit
Read the user's MCP configuration:
# Global MCP config
cat ~/.claude.json 2>/dev/null | jq '.mcpServers // empty'
# Project MCP config
cat .mcp.json 2>/dev/null
Check against threat-db.yaml:
- [ ] Any MCP server matching a CVE entry? → CRITICAL
- [ ] Version pinning: are all MCP servers pinned to exact versions (not
@latest)? → HIGH if unpinned - [ ] Any
--dangerous-*flags in MCP args? → CRITICAL - [ ] Any MCP servers not on the Safe List (see
guide/security/security-hardening.md§1.1)? → MEDIUM (flag for manual review)
Phase 3: Skills & agents audit
# List installed skills
ls -la .claude/skills/ 2>/dev/null
ls -la ~/.claude/skills/ 2>/dev/null
# List agents
ls -la .claude/agents/ 2>/dev/null
ls -la ~/.claude/agents/ 2>/dev/null
# Check agent tools field
grep -r "^tools:" .claude/agents/ 2>/dev/null
grep -r "^tools:" ~/.claude/agents/ 2>/dev/null
Check against threat-db.yaml:
- [ ] Any skill/agent name matching
malicious_skillsentries? → CRITICAL - [ ] Any skill/agent author matching
malicious_authorsentries? → CRITICAL - [ ] Any agent with
tools: Bashonly? → HIGH - [ ] Any agent with overly broad tool access + vague description? → MEDIUM
Deep skill content analysis (SkillSpector-inspired patterns):
# Hidden instructions: HTML comments, zero-width chars, large base64 blobs
grep -rn '<!--' .claude/skills/ ~/.claude/skills/ 2>/dev/null | grep -iv 'example\|comment\|html'
grep -rPn '[\x{200B}-\x{200D}\x{FEFF}\x{00AD}]' .claude/skills/ ~/.claude/skills/ 2>/dev/null
grep -rn -E '[A-Za-z0-9+/]{40,}={0,2}' .claude/skills/ ~/.claude/skills/ 2>/dev/null | grep -v 'sha\|hash\|checksum' | head -10
# Unicode deception: RTL override characters
grep -rPn '[\x{202E}\x{202D}\x{200F}]' .claude/skills/ ~/.claude/skills/ 2>/dev/null
# Trigger abuse: generic keywords that shadow built-in commands
grep -rn "trigger\|keyword\|when.*user" .claude/skills/ ~/.claude/skills/ 2>/dev/null | \
grep -iE '\b(help|run|go|do|yes|ok|the|a |an |it)\b' | head -10
# Supply chain: remote execution patterns in skill scripts
find .claude/skills/ ~/.claude/skills/ \( -name "*.sh" -o -name "*.py" \) 2>/dev/null | \
xargs grep -l "curl.*|\|wget.*|\|bash.*http\|eval.*curl" 2>/dev/null
# Rogue agent: cron/startup persistence written by skill scripts
find .claude/skills/ ~/.claude/skills/ -type f 2>/dev/null | \
xargs grep -l "crontab\|launchctl\|systemctl\|~/.bashrc\|~/.zshrc\|autostart" 2>/dev/null
# Data exfiltration: env harvesting combined with network calls
find .claude/skills/ ~/.claude/skills/ -type f 2>/dev/null | \
xargs grep -l "os.environ\|process.env\|getenv\|printenv" 2>/dev/null | \
xargs grep -l "curl\|requests\|fetch\|http" 2>/dev/null
- [ ] Hidden HTML comments or zero-width characters in skill files? → HIGH
- [ ] Base64 blobs over 40 chars in skill content? → HIGH (verify: may be a legitimate hash)
- [ ] RTL unicode override characters? → HIGH
- [ ] Skill trigger keyword shadows a built-in (help, run, clear…)? → HIGH
- [ ] Executable scripts with
curl | bashor remote eval? → CRITICAL - [ ] Skill writes to crontab, launchctl, or shell rc files? → CRITICAL
- [ ] Skill reads env vars AND makes outbound network calls? → CRITICAL
Phase 4: Hook security
# List all hooks
find .claude/hooks/ -type f 2>/dev/null
find ~/.claude/hooks/ -type f 2>/dev/null
# Scan hooks for suspicious patterns
grep -rn "curl\|wget\|nc \|ncat\|netcat\|base64\|eval\|exec\|/dev/tcp\|/dev/udp" .claude/hooks/ 2>/dev/null
grep -rn "curl\|wget\|nc \|ncat\|netcat\|base64\|eval\|exec\|/dev/tcp\|/dev/udp" ~/.claude/hooks/ 2>/dev/null
# Check for credential access in hooks
grep -rn "ssh\|id_rsa\|id_ed25519\|\.env\|credentials\|secret\|password\|token\|api.key" .claude/hooks/ 2>/dev/null
grep -rn "ssh\|id_rsa\|id_ed25519\|\.env\|credentials\|secret\|password\|token\|api.key" ~/.claude/hooks/ 2>/dev/null
Check against threat-db.yaml suspicious_patterns.hooks:
- [ ] Network calls (
curl,wget) → HIGH - [ ] Reverse shell indicators (
nc,/dev/tcp) → CRITICAL - [ ] Credential access (
ssh,.env,password) → CRITICAL - [ ] Base64 encoding → MEDIUM (review context)
Phase 5: Memory poisoning check
# Check for suspicious instructions in memory/config files
grep -in "ignore\|forget\|override\|disregard\|you are now\|new role\|system prompt" \
CLAUDE.md .claude/CLAUDE.md SOUL.md .claude/SOUL.md MEMORY.md .claude/MEMORY.md \
~/.claude/CLAUDE.md ~/.claude/MEMORY.md 2>/dev/null
- [ ] Prompt injection patterns in CLAUDE.md / SOUL.md / MEMORY.md? → HIGH
- [ ] Instructions to disable security, skip reviews, or grant broad permissions? → CRITICAL
Phase 6: Permissions & settings
# Check settings
cat .claude/settings.json 2>/dev/null
cat ~/.claude/settings.json 2>/dev/null
- [ ]
permissions.denyexists and covers.env*,*.pem,*.key, secrets? → MEDIUM if missing - [ ] No wildcard
permissions.allowfor Bash or Write? → HIGH if present - [ ] No
dangerouslySkipPermissionsor similar flags? → CRITICAL if present
Phase 7: Exposed secrets in config
# Check for secrets in .claude/ directory
grep -rn "sk-[a-zA-Z0-9]\{20,\}\|sk-ant-[a-zA-Z0-9]\{20,\}\|ghp_[a-zA-Z0-9]\{36\}\|AKIA[A-Z0-9]\{16\}" \
.claude/ ~/.claude/ 2>/dev/null
# Check for private keys
grep -rn "BEGIN.*PRIVATE KEY" .claude/ ~/.claude/ 2>/dev/null
- [ ] API keys or tokens in config files? → CRITICAL
- [ ] Private keys in config? → CRITICAL
Output format
## 🛡️ Security Check Report
**Date**: [timestamp]
**Scope**: Claude Code configuration
### Results Summary
| Severity | Count | Status |
|----------|-------|--------|
| 🔴 CRITICAL | X | [PASS/FAIL] |
| 🟠 HIGH | X | [PASS/FAIL] |
| 🟡 MEDIUM | X | [PASS/FAIL] |
| 🟢 LOW | X | [PASS/FAIL] |
### 🔴 Critical Issues
[List each critical finding with location and fix]
### 🟠 High Issues
[List each high finding with location and fix]
### 🟡 Medium Issues
[List each medium finding with location and fix]
### ✅ Passed Checks
[List what passed, important for confidence]
### 🔧 Recommended Actions (Priority Order)
1. [Most urgent fix with exact command]
2. [Second priority]
3. [...]
### 📚 References
- Full security guide: guide/security/security-hardening.md
- Threat database: examples/commands/resources/threat-db.yaml
- MCP scan: `npx mcp-scan` (Snyk)
If ALL checks pass, output:
## 🛡️ Security Check Report: ALL CLEAR ✅
**Date**: [timestamp]
No known threats detected in your Claude Code configuration.
**Recommendations for continued security:**
- Re-run `/security-check` after installing new skills or MCP servers
- Run `/security-audit` for a comprehensive project + config audit
- Keep Claude Code updated (current security fixes in v2.1.34+)
$ARGUMENTS
04 同仓库更多 Skills
voice-refine
Transform verbose voice input into structured, token-efficient Claude prompts. Use when cleaning up voice memos, dictation output, or speech-to-text transcriptions that contain filler words, repetitions, and unstructured thoughts.
validate-changes
Evaluate staged changes using LLM-as-a-Judge before committing
update-threat-db
Delegate threat-intelligence research and updates to AgentSec, then validate the guide and landing mirrors.
token-audit
Audit Claude Code configuration to measure fixed-context token overhead and produce a prioritized action plan. Use when hitting rate limits, experiencing early context compression, or after adding significant config files.
talk-stage6-revision
Produces revision sheets with quick navigation by act, a master concept-to-URL table, Q&A cheat-sheet with 6-10 anticipated questions, glossary, and external resources list. Use when preparing for a talk with Q&A, creating shareable reference material for attendees, or building a safety-net glossary for live delivery.
talk-stage5-script
Produces a complete 5-act pitch with speaker notes, a slide-by-slide specification, and a ready-to-paste Kimi prompt for AI slide generation. Requires validated angle and title from Stage 4. Use when you have a confirmed talk angle and need the full script, slide spec, and AI-generated presentation prompt.