ClawHub
Back to skill

Security audit

AI OS Blueprint

Security checks for vulnerabilities and agentic risk

Overview

This is a planning and checklist skill for building an AI workflow, with some automation and integration risks that are visible and purpose-aligned.

Installers should treat this as workflow architecture guidance, not a turnkey automation. Before connecting Slack, Notion, GitHub, or MCP servers, use least-privilege tokens, define what data may be sent externally, keep write actions behind approval gates, and avoid committing secrets when backing up configuration.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Credential Access

High
Category
Privilege Escalation
Content
|-----------|-------------|-------------|----------|------------|
| AI client installed (Claude Desktop / OpenClaw) | None | Old version | Current | ___ |
| MCP servers connected | 0 tools | 1-2 tools | 3+ tools | ___ |
| API keys stored securely | In chat history | In .env file | In secrets manager | ___ |
| Config file backed up | Never | Occasionally | Version controlled | ___ |
| Restart/reload procedure known | Unknown | Sometimes works | Documented | ___ |
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
|-----------|-------------|-------------|----------|------------|
| AI client installed (Claude Desktop / OpenClaw) | None | Old version | Current | ___ |
| MCP servers connected | 0 tools | 1-2 tools | 3+ tools | ___ |
| API keys stored securely | In chat history | In .env file | In secrets manager | ___ |
| Config file backed up | Never | Occasionally | Version controlled | ___ |
| Restart/reload procedure known | Unknown | Sometimes works | Documented | ___ |
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill explicitly promotes integrations that can post to Slack, write to Notion, and create GitHub issues, but it does not include a clear user-facing privacy/data handling warning about what information may be sent to external services. In this context, the skill is operational guidance rather than executable code, but encouraging automatic outbound writes without explicit disclosure increases the risk of unintended data exposure or transmission of sensitive context.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
□ At least 1 automated loop running
```

**Done when:** You get a Slack notification from your agent without asking for it.

---
Confidence
83% confidence
Finding
The phrase 'You get a Slack notification from your agent without asking for it' normalizes autonomous outbound actions. Although the broader document does mention approval gates for write operations, this success criterion still encourages unattended behavior that could send incorrect, sensitive, or premature information to external channels.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
| One-size-fits-all model | Burning credits on simple tasks | Apply Model Routing Matrix |
| Serial agent work | Tasks take 10x longer than needed | Enable sub-agent parallelism |
| Prompt not skill | Same prompts repeated every session | Package into SKILL.md |
| No approval gate | Agent sends external messages unexpectedly | Add gate to all write ops |
| Config not backed up | One system wipe loses everything | Git-commit your config |

---
Confidence
75% confidence
Finding
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Static analysis

No suspicious patterns detected.