autoresearch

name	autoresearch
description	Scaffold and run Karpathy-style autoresearch loops in any git repo. This skill should be used when setting up autonomous code improvement, generating adversarial eval harnesses, running hypothesis-implement-eval-keep/discard loops, or checking autoresearch progress. Triggers on "autoresearch", "autonomous improvement", "eval loop", "hypothesis loop", "self-improvement loop".
argument-hint	["init|eval-gen|run|status|resume"]

Five Invariants (never violate)

1. Single mutable surface — one hypothesis per iteration, one change per experiment
2. Fixed eval budget — eval runs in bounded time, no network calls in gates
3. One scalar metric — composite score drives keep/discard, not vibes
4. Binary keep/discard — improved = keep, else revert git reset --hard HEAD~1
5. Git-as-memory — every experiment is a commit, discards are reverts, history is the log

Safety rules

• Never modify .lab/ contents during hypothesis implementation
• Never skip eval — every commit must be evaluated before keep/discard
• Always revert on crash — atexit handler restores git state
• Runner uses subscription auth (claude -p with ANTHROPIC_API_KEY stripped)

Autoresearch

Scaffold and run autonomous code improvement loops in any git repo. The pattern: generate a hypothesis via claude -p, implement it, run programmatic eval gates, keep if the composite score improves, discard if it doesn't. Proven across 50+ iterations on two codebases (shadow-engine: 0.69 to 1.0, perplexity-clone: search quality optimization).