Mutator Docs

1. Install the CLI

pip install eas-mutator
mutator auth login <your-api-key>

2. Configure a run

# .mutator.yaml
name: api-service
repo: github.com/your-org/api-service
target_paths:
  - src/
generators:
  - rules      # deterministic rule-based fixes
  - search     # search-based refactors
  - llm        # LLM-assisted rewrites
validation:
  test_command: "npm test"
  coverage_delta_min: 0.0     # never lower coverage
  block_on_regression: true
  type_check: true
deploy:
  staging_first: true
  staging_soak_minutes: 60
  drift_tolerance: "2pct"     # behavior drift vs baseline
  rollout: [0.01, 0.10, 0.50, 1.00]
  auto_revert_on: ["error_rate", "health_check"]
  require_approval: true      # you own the prod-deploy decision

3. Run it

mutator run start .mutator.yaml
mutator run status     # poll candidates surviving each gate
mutator promote <candidate-id> --to=staging

Core concepts

• Generators — tiers of candidate-change producers (rule-based, search-based, LLM-assisted). Candidates compete; only those that pass your gates advance.
• Quality gates — your test suite plus coverage-delta, regression, and type-safety checks. A candidate that overfits the tests instead of genuinely fixing the issue is rejected.
• Concurrency lock — prevents concurrent edits from racing each other into a deploy.
• Closed-loop drift — staging behavior is continuously compared to the expected baseline. Drift beyond your tolerance quarantines the change.
• Gradual rollout — promoted changes ramp from 1% of traffic upward, advancing only after passing each tier's health and error-rate check, with auto-revert on breach.