Agents + Genome: The Full Picture

# Step 1: Check agent's current genome state
ace_genome_state({ agent_name: "guardian" })

# Step 2: Train with domain-relevant text (20 generations)
ace_genome_evolve({
  text: "Found contradiction: architecture doc says 'PostgreSQL only'
but decision #42 approves Redis for caching. Stale data in the
patterns collection — last updated 6 months ago. Knowledge graph
has 3 orphaned nodes with no relationships.",
  agent_name: "guardian",
  generations: 20
})

# Step 3: Verify traits shifted as expected
ace_genome_state({ agent_name: "guardian" })
# Expected: PRC ↑ (catches contradictions), REG ↑ (enforces rules)

# Step 4: The next time the Guardian runs, it uses
# its evolved genome for the observe/reason/act/reflect loop

# After agents have run for a while (days/weeks):
ace_genome_consolidate()

# Example output: The Guardian's elevated PRC (1.7) exceeds
# master (1.1) by >30%. Consolidation transfers a conservative
# 10% boost: master PRC becomes 1.16.
#
# Now ALL agents benefit from the Guardian's heightened perception
# without losing their own specialization.

# Compare Guardian vs Risk Engine genomes
ace_genome_state({ agent_name: "guardian" })
ace_genome_state({ agent_name: "risk_engine" })

# Expected divergence:
# Guardian: PRC 1.7, REG 1.6, RSN 1.2 (perceives + regulates)
# Risk:     RSN 1.8, INT 1.5, PRC 1.3 (reasons + integrates)
#
# Divergence proves genuine specialization, not random drift.

# Visual comparison available in Dashboard → Genome Lab
# Shows radar charts side by side for any two agents

# Export the Guardian's trained genome
ace_genome_export({ agent_name: "guardian" })
# → { dna: "eyJnZW5lcyI6...", fingerprint: "a1b2c3..." }

# Switch to a new namespace
ace_set_namespace({ namespace: "new-project" })

# Import the trained genome
ace_genome_import({
  dna: "eyJnZW5lcyI6...",
  fingerprint: "a1b2c3..."
})

# The new project now has a Guardian with 6 months of
# evolved perception — no training period needed.