backend-hardening

backend-hardening

Audits a backend or API for security vulnerabilities, stability bugs, and design flaws, explains each one in plain language, then fixes them on request.

Built for developers who are strong at frontend and learning backend — including people whose backend was largely written by an AI. It reports before it edits, and every finding comes with a concrete consequence and a way to check it yourself.

What it covers

Stack-specific checks for Node/TypeScript (Express, Nest, Fastify, Hono, Next.js), Python (Django, DRF, FastAPI, Flask), PHP (Laravel, Symfony), Ruby (Rails, Sinatra), Go, Spring Boot, ASP.NET Core, and Supabase/Firebase/Appwrite/PocketBase.

Install

npx skills add https://qala.lol/nuu-maan/backend-hardening

Works in any agent that supports the Agent Skills standard — Claude Code, Cursor, Codex, Copilot, Gemini CLI, OpenCode, Goose, and others.

Or, in Claude Code, install the whole collection as a plugin:

/plugin marketplace add Nuu-maan/skill
/plugin install qala@qala

Use

Ask for it in your own words — the skill triggers on requests like:

• "Review my backend before I deploy"
• "Is my API secure?"
• "Check this for vulnerabilities"
• "I built this with AI, what's wrong with it?"

Or invoke it directly in Claude Code:

/backend-hardening

You can scope it: /backend-hardening focus on the auth routes or /backend-hardening just check for exposed secrets.

What to expect

1. It identifies your stack from your files.
2. It runs a fast sweep of the twelve highest-signal checks.
3. It reads the relevant deep-dive checklists for what your code actually does.
4. It verifies each candidate finding against the real code path before reporting it.
5. You get a severity-ranked report with file:line evidence and a fixed-code example per finding.
6. It offers to fix, in priority order. It will not edit anything until you say yes.

If it finds nothing serious it says so, rather than padding the report.

Claude Code variant

The canonical SKILL.md uses only the six fields in the Agent Skills spec, so it works everywhere. If you use it exclusively in Claude Code, these frontmatter additions are worth considering:

# Run the audit in an isolated subagent so the (long) reading phase does not
# consume your main conversation's context.
context: fork
agent: Explore
background: false

Trade-off: with context: fork the audit runs without your conversation history, and the fix phase has to happen back in the main session. Good for large codebases, unnecessary for small ones.

# Pre-approve read-only tools plus the ecosystem audit commands, so the audit
# runs without permission prompts.
allowed-tools: Read, Grep, Glob, Bash(npm audit*), Bash(pip-audit*), Bash(govulncheck*), Bash(bundle audit*)

# Load automatically only when working in server-side code.
paths: ["**/api/**", "**/routes/**", "**/controllers/**", "**/handlers/**", "**/server/**"]

Do not add these fields to a copy you upload to claude.ai or the Skills API — those paths reject non-spec keys with a hard error.

Limitations

• Static review. It reads code; it does not run your app or attack it. Findings that depend on runtime configuration (environment variables set in a hosting dashboard, dashboard-configured database rules, network exposure) are flagged as unverifiable rather than guessed at.
• It is not a substitute for a penetration test on anything handling money, health data, or credentials at scale.
• Coverage is broad rather than exhaustive. It targets the defects that actually appear in small-team and AI-assisted backends.

Files

backend-hardening/
├── SKILL.md                        # entry point: workflow, severity rubric, routing
├── references/                     # loaded only when relevant
│   ├── authentication.md
│   ├── authorization.md
│   ├── input-validation.md
│   ├── data-layer.md
│   ├── api-design.md
│   ├── errors-and-resilience.md
│   ├── secrets-and-config.md
│   ├── abuse-and-limits.md
│   ├── operations.md
│   └── fixing-safely.md
├── stacks/                         # one is loaded, matching the project
│   ├── node.md
│   ├── python.md
│   ├── php.md
│   ├── ruby.md
│   ├── go.md
│   ├── jvm-dotnet.md
│   └── baas.md
└── templates/
    └── report.md

Only SKILL.md enters context when the skill triggers. Everything else is read on demand, so the depth costs nothing on a small audit.

License

MIT