DEV Community

Mehmet TURAÇ
Mehmet TURAÇ

Posted on • Originally published at github.com

Stop Your AI Agent From Building Tools That Already Exist

#claudecode #ai #productivity #programming

Your agent just wrote a custom PDF parser.

There were four maintained libraries that do exactly this. It didn't check. It never does.

This is the default behavior of every coding agent: task arrives → code is written → you maintain the bespoke solution forever.

skill-hunter is the missing pause.

What It Does

skill-hunter is a pre-execution layer for coding and automation agents. Before your agent writes a single line, it:

  1. Classifies the request — what kind of task is this?
  2. Scans the ecosystem — MCP servers, CLIs, npm/pip packages, APIs, GitHub repos, existing repo utilities
  3. Scores candidates — fit, maintenance activity, permissions, security, docs quality, license, integration effort
  4. Recommends a path — reuse, adapt, build minimally, or build custom
  5. Gates risky actions — installs, credentials, external service connections, destructive ops require explicit approval

The agent checks the toolbox before building another hammer.

The Before / After

Without skill-hunter:

User: "Parse these PDFs and extract invoices."
Agent: writes 200-line custom parser

With skill-hunter:

User: "Parse these PDFs and extract invoices."
Agent: checks PDF libraries, OCR tools, invoice extraction APIs, asks whether accuracy, cost, privacy, or offline processing matters
Agent: "pdfplumber covers 90% of this. Want me to wrap it with your schema instead?"

Install in 30 Seconds

Claude Code:

/plugin marketplace add mturac/skill-hunter
/plugin install skill-hunter@skill-hunter
Enter fullscreen mode Exit fullscreen mode

Codex CLI:

codex plugin marketplace add mturac/skill-hunter
Enter fullscreen mode Exit fullscreen mode

Then in ~/.codex/config.toml:

[plugins."skill-hunter@skill-hunter"]
enabled = true
Enter fullscreen mode Exit fullscreen mode

OpenClaw:

/openclaw-skills:skill_hunter
Enter fullscreen mode Exit fullscreen mode

Why This Matters

The problem isn't that agents write bad code. The problem is that agents write unnecessary code — and then you're stuck maintaining it.

Every custom solution is a maintenance debt. Every dependency you don't introduce is a security surface you don't expose. Every API you don't reinvent is a battle-tested edge-case handler you get for free.

skill-hunter doesn't stop agents from building. It stops agents from building what already exists.

GitHub: mturac/skill-hunter

If this solves a real pain for you, a star helps me know what to keep building.

Top comments (0)