Lesson 3: Continuous Evolution Workflow

The Evolution Loop

From v0.42.0, Sruja provides an evolutionary workflow that continuously monitors and improves architecture:

┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│ Instrumentation│ → │ Context Hints│ → │Infra Discovery│
└─────────────┘    └─────────────┘    └─────────────┘
        ↓                ↓                  ↓
   Measure health    Suggest improvements  Auto-optimize

Maturation Phases

Phase 1: Instrumentation

Start by measuring what you have:

sruja discover -r . --context

This builds the initial architecture graph.

Phase 2: Context Hints

Add human knowledge to guide evolution:

sruja intent propose -r . --from-diff

Phase 3: Infrastructure Discovery

Let Sruja find infrastructure patterns:

sruja scan -r . --infra

The Evolution Command

# Run evolution analysis
sruja evolution -r .

# Show recommended changes
sruja evolution -r . --propose

# Apply recommended changes
sruja evolution -r . --fix

Drift Detection and Resolution

# Detect drift
sruja drift -r .

# Auto-fix where possible
sruja drift -r . --fix

# Show blast radius of changes
sruja impact <component> -r . --depth 3

Agentic memory and bounded agent loop

Sruja records architecture-bounded learnings in .sruja/agent_memory.json and supports a review-before-apply agent loop. Sruja is a harness—not a full autonomous coding agent; your editor or CI hosts the LLM.

# Emit a reviewable plan grounded in repo evidence
sruja agent plan -r . --goal "Reduce drift on API layer" --file src/api/main.rs --print

# After human or CI review, apply the plan
sruja agent apply -r . --plan docs/plans/<run-id>.json

# Optional: parallel sandbox trajectories (MaTTS) when worktrees are available
sruja agent run -r . --goal "..." --file src/api/main.rs --mode apply --trajectories 3

# Inspect and curate learnings
sruja agent history -r .
sruja agent curate -r .

Enable automatic learning capture in .sruja/config.toml:

[agent]
auto_record_learnings = true

See Grounded harness and continual learning for the full host-vs-harness model.

Integrating into CI/CD

# .github/workflows/evolution.yml
- name: Check Architecture Health
  run: |
    sruja health -r . --format json > health.json
    sruja drift -r . --fail-on-drift

- name: Evolution Check (daily)
  if: github.event_name == 'schedule'
  run: sruja evolution -r . --propose

Summary

The continuous evolution workflow:

1. Instrument: Build architecture graph
2. Measure: Run health checks
3. Detect: Find drift with sruja drift
4. Propose: Get evolution suggestions
5. Apply: Use sruja evolution --fix
6. Learn: Agentic memory improves over time

Module Complete!

You've completed the Evolutionary Architecture module. You now understand:

• ✅ Fitness functions and how to define them
• ✅ Health scoring and community detection
• ✅ Continuous evolution workflow with maturation phases