ADR 0010: Govern autofix behavior with safety-first defaults

• Status: Accepted
• Date: 2026-02-28

Context

This plugin includes rules where some transformations are fully safe while others are report-only.

Autofix behavior must remain conservative and deterministic. If a rewrite is not provably safe in all supported contexts, the rule should report (or suggest) instead of automatically fixing.

Decision

Adopt a formal autofix governance model:

1. Rules should only emit fix when safety is deterministic.
2. Rules should emit suggest for behavior-sensitive migrations.
3. Rule docs should specify whether a rule is report-only, suggestion-capable, or autofixable.

Rationale

1. Operational safety: automated rewrites must preserve semantics.
2. Predictable rollout: teams can start with diagnostics and adopt fixes gradually.
3. Consistent authoring: rule authors follow one shared safety bar for fixes.

Consequences

• Fix behavior is intentionally policy-driven, not purely rule-local.
• Rule authors must classify rewrites as deterministic (fix) vs contextual (suggest/report-only).
• Migration playbooks can stage adoption by enabling presets in phases.

Revisit Triggers

Re-evaluate if:

• ESLint introduces stronger first-class fix governance primitives,
• the plugin requires finer-grained per-rule fix policy controls,
• or contributors report repeated unsafe-fix edge cases.