Agent skill
refactor-break-bw-compat
Refactor by removing backward compatibility and legacy layers. Use when modernizing APIs, cleaning up migration debt, removing compat shims, or eliminating stale feature flags.
Install this agent skill to your Project
npx add-skill https://github.com/OutlineDriven/odin-claude-plugin/tree/main/skills/refactor-break-bw-compat
SKILL.md
Breaking Refactors — Freedom Through Structure
Compatibility layers are coupling debt with interest. Every compat shim is a decision deferred, not a decision avoided. Break cleanly, break once, break with evidence.
Principles
- No half-measures. A partial migration is worse than no migration — it doubles the surface area and confuses every reader.
- One migration direction. Old-to-new only. Never add new-to-old adapters; that entrenches the old path.
- Blast radius awareness. Map every consumer before removing anything. Surprise breakage is a planning failure, not a courage signal.
- Dead code is a lie. "Just in case" code is not dead — it actively misleads readers about what the system does.
- Compat shims are temporary. If a shim has no removal date, it is permanent. If it is permanent, it is architecture. Decide which.
Reconnaissance Checklist
Before breaking anything, find and catalog:
@deprecated/#[deprecated]/warnings.warnmarkers — especially ones without removal versions- Version-gated code paths (
if version >= X, feature flags,#[cfg(feature = "legacy")]) - Adapter / shim / bridge / wrapper layers that translate between old and new interfaces
- Dual serialization formats (v1/v2 JSON schemas, protobuf
oneofwith legacy fields) - Tests that exist solely to validate backward-compatible behavior
- Configuration keys that toggle between old and new behavior
- Re-export / forwarding modules that alias old paths to new locations
- Changelog entries promising deprecation timelines
Decision: Break or Not?
| Signal | Break | Do NOT Break |
|---|---|---|
| Zero external consumers | Yes | — |
| Single internal consumer, you own it | Yes | — |
| Well-tested, high coverage | Yes | — |
| Clear new path exists | Yes | — |
| External/public API with unknown consumers | — | Not without migration plan |
| No tests covering the boundary | — | Write tests first, then break |
| Multiple consumers, unclear ownership | — | Map consumers first |
| Compat layer under active use by migration-in-progress | — | Finish migration first |
Execution Strategy
- Map blast radius. List every file, module, and external consumer that references the old API. Use
ast-grep,rg, or equivalent — not guesswork. - Snapshot current behavior. Ensure tests cover the old path. If coverage is missing, add characterization tests before removal.
- Remove the old path. Delete the compat layer, adapter, or legacy code. Do not comment it out.
- Update all call sites. Migrate every reference found in step 1 to the new API. Compile/typecheck after each batch.
- Delete orphaned tests. Tests that validated the old path are now dead weight. Remove them.
- Search for ghosts. Grep for string references, config keys, environment variables, documentation links, and error messages that mention the old API.
- Verify no dead imports/deps. Check for unused imports, packages, or dependencies that only the old path required.
Anti-patterns
- "Just in case" paths — keeping old code "in case someone needs it." That is what version control is for.
- Partial migration — half the codebase on new API, half on old. Worse than either alone.
- Commenting out instead of deleting — commented code is invisible debt that greps cannot find.
- Compat-of-compat — wrapping a compat layer in another compat layer. Two wrongs do not make an abstraction.
- Deprecated without removal date — a deprecation warning without a deadline is a suggestion, not a plan.
- Stale feature flags — flags that are always on (or always off) in every environment. Delete the flag, keep the winning path.
Validation Gates
| Gate | Condition |
|---|---|
| Blast radius mapped | Every consumer of old API identified and listed |
| Tests green pre-removal | Existing tests pass before any deletion begins |
| Zero references post-removal | ast-grep / rg for old API names returns zero hits |
| No dead imports/deps | No unused imports, packages, or type declarations remain |
Exit Codes
| Code | Meaning |
|---|---|
| 0 | Clean break — old API fully removed, all consumers migrated, tests pass |
| 1 | Partial — old references remain in code, docs, or config |
| 2 | Tests broken — removal caused test failures not yet resolved |
| 3 | External consumers found — need migration plan before proceeding |
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
pr-merge-temporal
Merge multiple PRs into a temporal integration branch before merging to base, with ordered conflict resolution. Use when you want to validate a set of PRs together on a staging branch before advancing the base branch.
tests-adversarial
Write adversarial tests that intentionally stress failure paths. Use when hardening error handling, stress-testing assumptions, validating boundary behavior, or hunting silent failures.
srgn-cli
Practical guide for building safe, syntax-aware srgn CLI commands for source-code search and transformation. Use when users ask for srgn commands, scoped refactors (comments/docstrings/imports/functions), multi-file rewrites with --glob, custom tree-sitter query usage, or CI-style checks with --fail-any/--fail-none.
askme
Verbalized Sampling (VS) protocol for deep intent exploration before planning. Use when starting ambiguous or complex tasks, when multiple interpretations exist, or when you need to explore diverse intent hypotheses and ask maximum clarifying questions before committing to an approach.
pr-merge-base
Merge one or more PRs into the base branch with queue-like sequencing and conflict resolution. Use when merging PRs that may conflict with each other or the base, requiring ordered application and intelligent conflict handling.
gh-fix-ci
Inspect GitHub PR checks with gh, pull failing GitHub Actions logs, summarize failure context, then create a fix plan and implement after user approval. Use when a user asks to debug or fix failing PR CI/CD checks on GitHub Actions and wants a plan + code changes; for external checks (e.g., Buildkite), only report the details URL and mark them out of scope.
Didn't find tool you were looking for?