Research better approaches, generate SKILL.md variations, score them, and update via PR.
npx skills add https://github.com/anajuliabit/aeon --skill autoresearch-anajuliabit
Please help me install this Agent Skill. Skill: Autoresearch Source: https://github.com/anajuliabit/aeon/tree/main/skills/autoresearch Command: npx skills add https://github.com/anajuliabit/aeon --skill autoresearch-anajuliabit
Autoresearch reduces skill rot by automatically improving an existing skill through targeted research, generating multiple improved versions, and selecting the best option without manual trial-and-error.
Run autoresearch with var set to the skill you want improved, for example: "Evolve the token-movers skill by running autoresearch with var=token-movers and returning the PR for the best variation."
High-intent search queries and answers about installing and using this skill.
Automatically improve an existing AI agent skill by researching better approaches, generating multiple SKILL.md variations, and scoring them against a weighted rubric to select the best version.
Prevent skill rot in automated prompt workflows by running a continuous improvement loop that evaluates target skills against new approaches and applies safe updates only when variations outperform the baseline.
Generate and test multiple variations of a markdown skill file by producing four distinct alternatives, scoring each against explicit rubric criteria, and submitting the winning version via a GitHub PR.
Automatically update GitHub PRs with better performing AI agent skills by applying rubric-based selection, preserving required frontmatter conventions, and skipping the update when no improvement over the baseline is found.
This self-healing automation works without manual trial-and-error by evaluating four variations across better inputs, sharper outputs, robust error handling, and a fundamentally different methodology to ensure safe skill evolution.
A limitation of using rubric scoring for skill evolution is that it strictly avoids regressions by skipping updates when an improvement is not found, meaning no PR is created if variations fail to beat the baseline.
yogogml0035-cpuAnalyze codebases to generate or optimize AGENTS.md project rules.
swarm-ai-researchGenerate four skill variations, score them, and commit the winner as a PR.
jlowinDiscover and invoke MCP tools from the command line.
whichguyScore prompts with the Prompt Critique Rubric and produce prioritized rewrites.
