# Agent Security Hardening (agent-security-hardening)

> Hardens the boundaries where an agent turns untrusted data into filesystem, network, memory or tool actions, producing concrete controls and tests.

- Category: [Hardening](https://www.myskills.tools/en/categories/hardening/)
- Source: Community · Affaan Mustafa
- Repository: https://github.com/affaan-m/ECC/tree/main/skills/agent-security-hardening (commit 9e951e2)
- SKILL.md: https://github.com/affaan-m/ECC/blob/main/skills/agent-security-hardening/SKILL.md
- Install command: `npx skills add affaan-m/ECC --skill agent-security-hardening`
- License: MIT
- Compatible with: Claude Code, Codex, Cursor, Copilot, OpenCode
- Permissions: read-repo ✓, write-files ✓, shell ✓, network ✗
- Status: Scan passed
- A rule scan cannot catch every risk. Read this skill's SKILL.md and check its permissions before installing.

## Core rule

Treat model output, retrieved content, tool output, memory and cross-agent messages as untrusted input.

## Execution Phases

1. **Map trust boundaries** — List every boundary as source -> parser -> validator -> side effect, recording the authorizing identity, validator, size limits, audit evidence and recovery path.
2. **Validate IDs, env and URLs** — Allowlist identifiers and reject empty env values; parse URLs, block private destinations, and stop DNS rebinding and cross-origin credential forwarding.
3. **Contain filesystem access** — Reject absolute paths, traversal and Windows-unsafe components, prove the resolved path stays in the workspace, and handle symlink races separately.
4. **Protect secrets and output** — Create sensitive temp files owner-only and keep secrets out of arguments, logs and model context; read tool output with a byte limit and return typed failures.
5. **Verification gate** — Verify six behaviors through the real entry point and report the command, exit status, rejected input and the observed absence of side effects.

## Permissions (summarised from SKILL.md)

| Capability | Needed | Details |
|---|---|---|
| File writes | Required | Writes validation, containment and other control code plus tests for the framework under review. |
| Command execution | Tests only | Runs tests through the real entry point to prove the hardened behavior and reports commands and exit status. |
| Network outbound | Not required | The skill itself makes no network calls; the network topics are outbound controls to be hardened. |

## SKILL.md frontmatter (YAML)

```yaml
---
name: agent-security-hardening
description: Security hardening guidance for AI agent frameworks that process untrusted content, invoke tools, write workspace files, manage runtime identifiers, or handle credentials. Use when building or reviewing an agent runtime, autonomous worker, tool gateway, memory service, or multi-tenant agent deployment. Do not use for general web application security or offensive testing.
metadata:
  origin: ECC
---
```

Home: https://www.myskills.tools/en/ · URL: https://www.myskills.tools/en/skills/agent-security-hardening/