skill-iteratelisted

# /skill-iterate This skill ports the serial hill-climb pattern from [`<workspace>/primatives/autoresearch/program.md`](../../../primatives/autoresearch/program.md) to skill iteration. Three load-bearing rules carry over verbatim: - **NEVER STOP mid-loop**: no mid-run halts inside a per-skill iteration loop — the loop drains to its budget cap or crashes; nothing else. - **Simplicity bias on ties**: a mutation that deletes code with a ≥0 score delta is a keep. - **Wall-clock-bounded experiments**: every per-iteration scoring call has a hard timeout; every per-skill loop has a hard wall-clock cap. Treat any deviation as a defect. --- ## When to use - The operator has 1+ scorable skills (i.e. `evals/evals.json` + `evals/test_scenarios.json` exist) and wants overnight hill-climb iteration. - Default invocation drains all scorable skills serially, one worktree per skill, budget-capped. - Use after `/skill-eval-setup` has bootstrapped evals for the target. ## When NOT to use - No evals present for the target skill — run `/skill-eval-setup <skill>` first. - Skill is in the embedded skip-list below (output is side-effect-driven, not text-gradable). - Skill is at a hill-climb plateau and needs radical exploration — use `/skill-evolve` (parallel A/B with operator-curated variants) instead. - Single-skill interactive iteration where the operator wants to review each diff — use `/build-step` against a hand-edited SKILL.md instead. --- ## Defaults | Flag | Default | Notes | |--