skill-evolvelisted

A/B-test N variant mutations of a skill in parallel. Takes --skill <name> and either --variants <file> (one strategy per line) or no variants (brainstorm mode — LLM proposes a candidate list and exits for operator curation, no worktrees created). When variants are supplied, creates N worktrees, dispatches N parallel agents that each apply their strategy and run the existing produce-grade loop, then produces a comparison table. The winner's branch is pushed and a `gh pr create` command is printed for operator review (no auto-PR). Loser-analysis is written under docs/investigations/skill-evolve/ for the next generation. Invoke as "/skill-evolve --skill <name> [--variants <path-or-inline>]".
aberson/skill-mesh · ★ 2 · AI & Automation · score 71

Install: claude install-skill aberson/skill-mesh

# Skill Evolve `/skill-evolve` runs N parallel A/B variants of a skill's SKILL.md against the skill's existing eval suite and reports which variant won. Built on top of the produce-grade loop documented in [`skill-eval-setup`](../skill-eval-setup/SKILL.md). The skill's whole reason to exist is to compress what would be N sequential `/skill-eval-setup` self-improvement runs into one parallel fanout. --- ## Execution model — autonomous, parallel-fanout, no auto-PR (HEAVY) This skill exists so the operator can test 3-5 candidate prompt mutations in one shot instead of running them sequentially over days. The whole point is parallel evaluation with honest reporting. Therefore: 1. **Parallel by default, via the scoring workflow.** Variants are evaluated by the shared workflow at [`../_shared/score_skill.workflow.js`](../_shared/score_skill.workflow.js), which fans out all variants — and within each variant, all scenarios — concurrently. Sequential variant runs defeat the whole purpose. If the Workflow tool is unavailable in this environment, halt with a clear error rather than silently falling back to a single self-grading agent (the defect this skill was rebuilt to avoid — see #6). 2. **No auto-PR.** Per workspace convention (never push without explicit ask, direct-to-master is the norm), this skill DOES NOT run `gh pr create`. It pushes the winner's branch to remote and PRINTS the `gh pr create` command for the operator to run after reviewing the diff locally. Auto-opening