# Security Audit

> Use when reviewing code for security vulnerabilities. Covers prompt injection, path traversal, command injection, and agent-specific attack vectors.

- Skill: `voidful/security-audit` (Agent Skill)
- Install (CLI): `npx skillmds@latest add voidful/security-audit`
- Raw SKILL.md: https://api.skillmd.com/api/skills/voidful/security-audit/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: Apache-2.0
- Author: voidful (https://skillmd.com/u/voidful)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/voidful/security-audit

---


# Security Audit

## Agent-Specific Threats

AI coding agents face unique security threats beyond traditional application security:

1. **Prompt injection via MEMORY.md / USER.md** — Malicious content written to memory files can alter agent behavior. Aixlarity's `memory_tool.rs` includes pattern scanning for this.
2. **Path traversal via tool calls** — Agent requests `read_file("../../etc/passwd")`. Aixlarity's trust system restricts this.
3. **Command injection via shell tool** — Agent constructs shell commands from untrusted input. The permission system is the primary defense.
4. **Exfiltration via fetch_url** — Agent sends sensitive data to external URLs. Sandbox policy controls this.

## Audit Checklist

For every code change, check:

- Does this introduce a new path where user input reaches a shell command?
- Does this bypass the trust / permission system?
- Does this write to MEMORY.md / USER.md without the safety scan?
- Does this expose API keys or credentials in output / logs?
- Does this fetch external URLs without sandbox policy check?

## In Aixlarity Source Code

Key security boundaries:
- `trust.rs` — Three-level trust model
- `agent/permissions.rs` — Permission prompt logic
- `tools/memory_tool.rs` — Memory safety scanning
- `tools/container.rs` — Container sandbox