Skip to content

AI Agent Skills

eigenhelm publishes agent skills for coding assistants. When an agent loads the skill, it gains a structured protocol for evaluating and improving code quality.

Install the skill

npx skills add metacogdev/skills --skill eigenhelm

Copy the skill to your project:

mkdir -p .opencode/skills/eigenhelm
# Copy SKILL.md from https://github.com/metacogdev/skills/tree/main/eigenhelm

mkdir -p .claude/skills/eigenhelm
# Copy SKILL.md

What the skill provides

The skill teaches agents to:

  1. Run eh evaluate --classify --format human <file> after writing code
  2. Interpret scores, decisions, and directives
  3. Follow an iteration protocol with strict limits to prevent infinite refinement loops

Iteration protocol

The skill enforces these rules:

Decision Action Max attempts
reject Address [high] directives, re-evaluate 3
marginal Address [high] directives if straightforward 2
accept Stop —

Stop rule: If the score doesn't improve by ≥ 0.03 between attempts, stop immediately.

Small file caveat: Files under ~80 lines will show [medium] improve_compression and [medium] review_structure — these are expected and should be ignored. Only PCA-derived directives (reduce_complexity, extract_repeated_logic) are actionable on small files.

Example agent prompt

You can also add eigenhelm to any agent by including this in your system prompt or project instructions:

After writing or modifying code, evaluate it with:
  eh evaluate --classify --format human <file>

If the decision is "reject", address [high] severity directives and re-evaluate (max 3 attempts).
If "marginal", address [high] directives if straightforward (max 2 attempts).
If "accept", move on.

Stop iterating if the score doesn't improve by 0.03 between attempts.