Convert document-oriented procedural skills into compact strategy gene control units.
npx skills add https://github.com/didi/mpx --skill skill2gene
Please help me install this Agent Skill. Skill: skill2gene Source: https://github.com/didi/mpx/tree/main/.agents/skills/skill2gene Command: npx skills add https://github.com/didi/mpx --skill skill2gene
Traditional procedural skills are document-oriented and consume around 2500 tokens, diluting the behavioral control signals that actually influence model reasoning. This Skill distills them into compact strategy genes of 200-300 tokens based on the Gene Evolution Protocol (GEP), maximizing control signal density per token.
Convert my existing procedural skill in the target directory into a set of compact strategy genes following the GEP protocol.
High-intent search queries and answers about installing and using this skill.
Read the entire skill package, identify control signals concentrated in workflow sections, split them by functional domain into atomic genes, then write each gene with keywords, a one-sentence summary, 3-7 strategy steps, and AVOID warnings. Finish by creating a GENES.md registry and validating coverage.
GEP defines a gene as an HTML comment metadata header (type, schema_version, id, signals_match) plus a strategy-gene tag body containing domain keywords, summary, and ordered strategy steps with embedded AVOID items. Each gene file targets 200-300 tokens.
Research shows genes are not merely shorter skills but a different abstraction: they organize experience as explicit control interfaces rather than documentation. Truncated skills underperform genes, and appending documentation back to genes actually degrades results.
Each gene should contain 200-300 tokens, which is the empirically validated optimal control budget. Exceeding this dilutes signal density, while falling below it may indicate missing decision constraints.
A single precise gene performs best according to the referenced research. Combining multiple genes, even nominally complementary ones, typically weakens control by splitting attention and blurring the control focus.
Encode failure patterns as standalone compact AVOID instructions placed at the end of a gene's strategy list. Independent failure warnings are the most effective encoding form, while mixing them into strategy steps weakens both.
nibojGenerate production-ready SKILL.md bodies from high-level skill concepts.
cipriantitireAnalyze runtime history and apply protocol-constrained patches to evolve AI agents.
charly-vibesCompare original and distilled prompts to verify lossless distillation.
sabyunrepoCompresses text by reducing token count through multiple optimization techniques.
