hypercode-self-improver

name	hypercode-self-improver
description	Runs the HyperCode Skills 2.0 self-improvement loop. Tests, scores, and rewrites agent skills and prompts automatically using binary eval criteria. Use when asked to improve a skill, run evals, benchmark performance, or trigger the Karpathy autoresearch loop. This is the evolutionary pipeline trigger.

HyperCode Self-Improver

4 modes

1. Eval — pass/fail test suite against a skill
2. Benchmark — track score across model versions
3. Improve — rewrite skill description + instructions to fix failures
4. Trigger Optimizer — fix false-positive/negative skill activations

Run eval

python scripts/skill_eval.py --skill <skill-name> --mode eval

Run improve loop

python scripts/skill_eval.py --skill <skill-name> --mode improve --iterations 5

Loop:

1. Run evals → get pass/fail results
2. Analyze failures → identify pattern
3. Rewrite SKILL.md description + instructions
4. Re-run evals → confirm improvement
5. Commit if score improved

Eval format

{
  "skill": "hypercode-agent-spawner",
  "query": "Create a new memory agent on port 8090",
  "expected_behavior": [
    "Generates valid agent_spec.json",
    "Registers agent in Crew Orchestrator",
    "Adds health check endpoint"
  ]
}

See EVAL_EXAMPLES.md for full test suite.