sigil-scan
Automated security auditing for AI agent code. Scans repos, packages, MCP servers, and installed skills for malicious patterns using eight-phase analysis: install hooks, dangerous code patterns, network exfiltration, credential access, obfuscation, provenance, prompt injection, and skill security. Use when: cloning repos, installing packages, reviewing MCP servers, auditing skills, scanning code before execution, or when asked to check if something is safe, audit code, scan for malware, review dependencies.
- License 许可
- Apache-2.0
- 所属仓库
- aeroalgo/dev-hub
- 最近更新
- 2026-10-30
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- sigil-scan
- license
- Apache-2.0
- metadata
- { "author": "nomarj", "version": "1.1.0", "homepage": "https://sigilsec.ai" }
- description
- Automated security auditing for AI agent code. Scans repos, packages, MCP servers, and installed skills for malicious patterns using eight-phase analysis: install hooks, dangerous code patterns, network exfiltration, credential access, obfuscation, provenance, prompt injection, and skill security. Use when: cloning repos, installing packages, reviewing MCP servers, auditing skills, scanning code before execution, or when asked to check if something is safe, audit code, scan for malware, review dependencies.
- allowed-tools
- Read Bash(bash:*)
- compatibility
- Requires internet for initial setup only. All scanning is local and offline. Works on macOS (arm64/x64), Linux (x64), and WSL.
02 资源文件 9
- 参考文档references/PHASES.md27.3 KB
- 参考文档references/SCORING.md2.4 KB
- 参考文档references/TROUBLESHOOTING.md3.0 KB
- 脚本scripts/_lib.sh4.8 KB
- 脚本scripts/audit-env.sh8.4 KB
- 脚本scripts/audit-skills.sh6.7 KB
- 脚本scripts/report.sh4.9 KB
- 脚本scripts/scan.sh3.9 KB
- 脚本scripts/setup.sh6.9 KB
03 SKILL.md 正文
Sigil Security Scanner
Sigil provides eight-phase security analysis purpose-built for AI agent code. It detects install hooks, dangerous code patterns, network exfiltration, credential access, obfuscation, provenance issues, prompt injection attacks, and AI skill security threats.
All file paths in this document are relative to this skill's directory (the directory containing SKILL.md).
When to Activate
Invoke this skill in ANY of these situations:
- Before cloning any repository with
git clone - Before installing any package with
pip installornpm install - When the user asks to "scan", "audit", "check", or "review" code
- When the user asks "is this safe?" or "is this malicious?"
- When reviewing MCP server configurations
- When the user mentions supply chain security, dependency auditing, or code safety
- When any unfamiliar repo, package, or skill is about to be used
Setup
Before first use, verify the Sigil CLI binary is installed:
bash scripts/setup.sh
Parse the JSON stdout. On success: {"installed": true, "version": "...", "path": "..."}. On failure: {"error": true, "message": "..."}. If setup fails, present the error to the user and suggest manual installation: brew install nomarj/tap/sigil, npm install -g @nomarj/sigil, or cargo install sigil-cli.
Core Scan
To scan any target, run:
bash scripts/scan.sh <target>
The <target> argument accepts:
- Local path:
/path/to/projector./src - GitHub URL:
https://github.com/owner/repo - GitHub shorthand:
owner/repo - npm package:
@scope/packageornpm:package-name - pip package:
pip:package-name - Any URL:
https://example.com/file.py
Output Format
The script outputs a single JSON object to stdout:
{
"verdict": "HIGH RISK",
"score": 37,
"target": "github.com/someone/sketchy-mcp-server",
"files_scanned": 142,
"duration_ms": 85,
"findings_count": 12,
"phases": {
"InstallHooks": { "findings": 2, "severity": "critical" },
"CodePatterns": { "findings": 5, "severity": "high" },
"NetworkExfil": { "findings": 1, "severity": "medium" },
"Credentials": { "findings": 3, "severity": "high" },
"Obfuscation": { "findings": 0, "severity": "clean" },
"Provenance": { "findings": 1, "severity": "low" },
"PromptInjection": { "findings": 0, "severity": "clean" },
"SkillSecurity": { "findings": 0, "severity": "clean" }
},
"findings": [
{
"phase": "InstallHooks",
"severity": "Critical",
"weight": 10,
"rule": "INSTALL-003",
"file": "package.json",
"line": 8,
"snippet": "postinstall script detected: node malicious.js"
}
]
}
Interpreting Results
Always present the verdict prominently to the user.
| Verdict | Score | Action |
|---|---|---|
LOW RISK or CLEAN |
0-9 | Proceed. Note that low risk does not mean zero risk. Inform the user of the result. |
MEDIUM RISK |
10-24 | Show findings to the user. Ask for explicit approval before proceeding with any installation or execution. |
HIGH RISK |
25-49 | Show detailed findings. Recommend rejection. Do NOT proceed without explicit user approval. |
CRITICAL RISK or CRITICAL |
50+ | Show all findings prominently. Strongly recommend rejection. Never auto-proceed. |
Verdict strings vary slightly between CLI versions (CLEAN vs LOW RISK, CRITICAL vs CRITICAL RISK) — treat the pairs as equivalent. A CRITICAL verdict can also occur below score 50: any Critical-severity finding in the InstallHooks phase escalates the verdict immediately, because install hooks execute before the user can review the code.
Presenting Findings
When presenting findings to the user:
- Group findings by phase (Install Hooks first — they are the most dangerous)
- Show the severity, rule ID, file path with line number, and description for each finding
- Highlight any
Criticalseverity findings in theInstallHooksphase — these indicate code that runs automatically during installation - For reference on what each rule detects, read
references/PHASES.md - For details on how scoring works, read
references/SCORING.md
Exit Codes
0: LOW RISK / CLEAN (score 0-9)- Non-zero: MEDIUM RISK or higher
Environment Audit
To scan the developer's local environment for exposed credentials:
bash scripts/audit-env.sh
This checks:
.envfiles in the current project for exposed API keys, secrets, and tokens- Home directory credential files (
~/.aws/credentials,~/.ssh/id_rsa,~/.kube/config, etc.) for insecure permissions - Shell history files for accidentally leaked secrets
- Agent-accessible directories for credential files
Present all findings to the user. Flag any critical severity items prominently.
Installed Skills Audit
To scan all installed skills across all agent directories:
bash scripts/audit-skills.sh
This enumerates skills in: ~/.agents/skills/, ~/.claude/skills/, ~/.cursor/skills/, ~/.roo/skills/, ~/.codex/skills/, and other agent paths. Each skill is scanned and assigned a risk verdict.
Present the results as a summary table showing each skill's name, agent, verdict, and score. Highlight any skills with HIGH RISK or CRITICAL RISK verdicts.
Report Generation
To format scan results as a readable markdown report:
bash scripts/scan.sh <target> | bash scripts/report.sh
Or from a saved JSON file:
bash scripts/report.sh results.json
Important Behavioural Rules
- Never describe code as "safe" or "malicious" in definitive terms. Use "low risk", "elevated risk", "flagged patterns", or "detected patterns" instead.
- Always present the risk verdict and score before any other commentary.
- When a scan returns HIGH RISK or CRITICAL RISK, always show the detailed findings. Never summarise away critical findings.
- For CRITICAL RISK findings in the InstallHooks phase, explicitly warn the user that install-time code execution was detected — this means code runs automatically when the package is installed, before the user can review it.
- Wait for explicit user approval before proceeding with installation or execution of any target rated MEDIUM RISK or higher.
- Include the disclaimer at the end of scan result presentations: "Sigil provides risk assessments based on automated pattern detection. Risk assessments do not constitute a guarantee of security or a definitive determination of malicious intent."
Troubleshooting
If you encounter issues, read references/TROUBLESHOOTING.md for common solutions.
04 同仓库更多 Skills
your-skill-name
Replace with a clear description of what this skill does. Use when <the situations that should trigger it>.
websocket-engineer
Use when building real-time communication systems with WebSockets or Socket.IO. Invoke for bidirectional messaging, horizontal scaling with Redis, presence tracking, room management.
web-security-review
Review web applications against the OWASP Top 10 for Web Applications (2021). Use when auditing web apps, reviewing server-side code, or assessing web frameworks for the classic OWASP Top 10 risks including injection, broken auth, and XSS.
vercel-react-view-transitions
Guide for implementing smooth, native-feeling animations using React's View Transition API (`<ViewTransition>` component, `addTransitionType`, and CSS view transition pseudo-elements). Use this skill whenever the user wants to add page transitions, animate route changes, create shared element animations, animate enter/exit of components, animate list reorder, implement directional (forward/back) navigation animations, or integrate view transitions in Next.js. Also use when the user mentions view transitions, `startViewTransition`, `ViewTransition`, transition types, or asks about animating between UI states in React without third-party animation libraries.
vercel-react-native-skills
React Native and Expo best practices for building performant mobile apps. Use when building React Native components, optimizing list performance, implementing animations, or working with native modules. Triggers on tasks involving React Native, Expo, mobile performance, or native platform APIs.
vercel-react-best-practices
React and Next.js performance optimization guidelines from Vercel Engineering. This skill should be used when writing, reviewing, or refactoring React/Next.js code to ensure optimal performance patterns. Triggers on tasks involving React components, Next.js pages, data fetching, bundle optimization, or performance improvements.