Skip to content

Configuration

Place .aion.yaml in the target repository root. AION reads it for repository scan defaults and for orchestration commands that resolve a repo_root.

Example

provider: openai
model: gpt-4.1
ignore_paths:
  - tests/*
  - scripts/generated_*.py
auto_repair_issue_types:
  - raw_sqlite_query
  - hardcoded_secret
  - missing_auth_decorator
auto_repair_min_confidence: 0.90
sandbox_mode: repository
sandbox_verification_commands:
  - python -m pytest tests/unit
auto_approve_verified_fixes: false
rollback_on_verification_failure: true

Fields

Field Type Default Description
provider string null Scan-time LLM provider, anthropic or openai
model string provider default Explicit model override for scan
ignore_paths list [] Glob patterns skipped during repository scanning
auto_repair_issue_types list built-in set Incident types allowed into automatic sandbox remediation
auto_repair_min_confidence float 0.85 Minimum incident confidence required before automatic remediation
sandbox_mode string repository file for single-file staging or repository for full repository staging
sandbox_verification_commands list [] Commands executed in the staged sandbox after built-in verification
auto_approve_verified_fixes boolean false Emit approved_for_rollout when staged verification passes
rollback_on_verification_failure boolean true Emit rollback instead of needs_human_review when staged verification fails

Resolution rules

  • CLI flags override matching .aion.yaml settings.
  • scan reads configuration from the target repository root.
  • process-event, process-event-queue, and inbox processing read configuration from the event repo_root.

Conservative staging:

auto_repair_min_confidence: 0.95
auto_approve_verified_fixes: false
rollback_on_verification_failure: false

Fast local experimentation:

sandbox_mode: file
auto_approve_verified_fixes: true
rollback_on_verification_failure: true

Operational notes

  • provider and model affect scan; deterministic repair and orchestration do not require an LLM.
  • sandbox_verification_commands execute inside the staged workspace and each command result is stored in the orchestration record.
  • Context extraction results are cached at ~/.aion-context.json.