Habib1001-m

Hakim — AI skill for Claude Code

AI community

A judgment layer for AI coding agents: smaller safe changes, native host integrations, and evidence-bound completion claims.

How to install Hakim

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Habib1001-m/hakim and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Hakim does

A judgment layer for AI coding agents: smaller safe changes, native host integrations, and evidence-bound completion claims.

Alternatives in AI

Academic Humanizer — Strip AI-writing tells from papers and grant proposals (NSF/NIH), while keeping scholarly voice and tying clai 1.1k ★
Agenttrail — Local observability map for AI coding agents 269 ★
Cupcake — A native policy enforcement layer for AI coding agents 255 ★

README

Hakim (حَكِيم)

**A judgment layer for AI coding agents.**

Smaller changes. Safer actions. Claims backed by evidence.

**Public beta · `v1.0.0-beta.11` · MIT · Codex · Claude Code · GitHub Copilot CLI · OpenCode**

Hakim gives capable coding agents a compact engineering decision model before they start adding code, dependencies, abstractions, or confident completion claims.

It does **not** turn your agent into a workflow bot. The host still owns permissions, tools, sandboxing, trust prompts, and execution. Hakim focuses on one narrower problem: **better judgment about what should change, how much evidence is enough, and what the result actually proves.**

Why Hakim exists

AI coding agents are increasingly good at implementation. They can still make expensive judgment mistakes:

adding a dependency when the platform already solves the problem;
creating an abstraction before there is a real need for one;
inspecting far more of the repository than the task requires;
weakening a security, migration, rollback, or compatibility guard to make a change look simpler;
declaring success from a green command that did not actually prove the user-visible result.

Hakim gives the agent a small decision ladder:

need?
  → reuse?
  → stdlib?
  → native platform?
  → accepted dependency?
  → smaller clear implementation?
  → minimum custom code

The goal is not the fewest lines. It is the **smallest sufficient safe change**.

What changes in practice

Without an explicit judgment layer, an agent may jump from a request directly to implementation.

With Hakim active, the agent is guided to answer a few higher-value questions first:

Do we need a change at all?
Can the repository or platform already do this?
What real guard must remain true?
What is the smallest coherent implementation?
What verification actually supports the complet