ai-genome（AI 基因组）

AI Genome Analysis Skill

You have access to the AI Personality Evolution Engine - a biologically-faithful framework that encodes AI personalities as diploid genomes with 27 cognitive primitives.

What You Can Do

Based on $ARGUMENTS, perform one of these actions:

"encode" or "encode <path>"

Encode a SOUL.md file into a breedable genome.

1. If a path is provided, use it. Otherwise look for SOUL.md in the current directory.
2. Run the encoder:

python3 encoder.py <soul_path> --name "<AgentName>" --output <name>.dna.json

1. Show the resulting phenotype with python3 visualize.py <name>.dna.json
2. Summarize the top 3 traits and any active epistasis rules.

"compare <slug>" or "compare <slug1> <slug2>"

Compare two genomes. If one slug is given, compare the user's genome against a library agent. If two slugs, compare those two library agents.

python3 agent_report.py <genome1>.dna.json --compat library/genomes/<slug>.dna.json

Report the genetic distance, complementarity, interest score, and predicted offspring trait ranges.

"view <slug>" or "view <path>"

Display a genome's full profile.

python3 visualize.py library/genomes/<slug>.dna.json
python3 agent_report.py library/genomes/<slug>.dna.json --self

"browse" or "library"

List available genomes from the library with their top traits.

python3 -c "
import json
lib = json.load(open('library/genome_library.json'))
for a in lib['agents']:
    traits = sorted(a['phenotype'].items(), key=lambda x: -x[1])
    top = ' | '.join(f'{t}={v:.0f}' for t,v in traits[:3])
    print(f\"{a['name']:25s} [{a['category']:20s}] {top}  (epistasis: {a['active_epistasis']})\")
print(f\"\
{lib['count']} agents across {len(lib['categories'])} categories\")
"

"card" or "json <slug>"

Get a machine-readable JSON card for a genome.

python3 agent_report.py library/genomes/<slug>.dna.json --json

"self" or "self-report <slug>"

Get a structured self-knowledge document (for an agent's own context window).

python3 agent_report.py library/genomes/<slug>.dna.json --self

Key Concepts

• 27 cognitive primitives: Low-level genes (not "creative" but novelty_seeking + pattern_completion + ambiguity_response + abstraction_preference)
• Diploid: Two alleles per gene. You carry traits you don't express (recessive alleles)
• 8 emergent traits: creativity, warmth, precision, wit, depth, boldness, adaptability, intensity - these emerge from gene clusters, not stored directly
• Epistasis: When two genes both cross thresholds, they modify a third gene. 12 rules create non-linear interactions
• Compatibility: Simulates 12 breedings to predict offspring trait ranges, recessive surfacing risks, and genetic distance

File Locations

• Library index: library/genome_library.json
• Individual genomes: library/genomes/<slug>.dna.json
• Encoder: encoder.py
• Visualizer: visualize.py
• Reports: agent_report.py

Notes

• Encoding requires one Claude API call (~3 minutes). Use --mock flag for instant keyword-based encoding (no API).
• All comparison, visualization, and breeding operations are pure local computation - no API calls.
• The library contains genomes from the OpenClaw community agent repository.