remove-code

Remove Code

Process

1. Understand the capability to remove (not just files): What user-facing or internal capability is being removed?

1. Trace dependencies before deleting anything:

• All references to the functionality across the codebase
• Code that ONLY exists to support this feature (can be fully deleted)
• Code that is SHARED with other features (must be preserved)
• Database migrations, config entries, env vars, feature flags, documentation that reference it

1. Present the trace: List everything to delete and everything to modify, grouped by file. Wait for user review.

1. Execute deletion:

• Fully dead files: delete
• Partially dead files: remove only dead parts
• Update imports, exports, index files
• Remove or update related tests

1. Verify:
2. Run the project's check command
3. Grep for remaining references to removed code (function names, class names, route paths, config keys) — there are always stragglers

Rules

• Describe the CAPABILITY being removed, not files — this finds things you forgot about
• Always verify with a full check run after deletion
• Do deletion in its own PR, separate from any new feature work
• Check .agents/common-mistakes/ for any removal-specific patterns before executing
• Big-output discipline. Heavy command output (project check, full git diff, repo-wide search, long log, large fetch) goes to /tmp/hawk-remove-code-<step>.log, then rg -n '<pattern>' /tmp/hawk-remove-code-<step>.log | head -50 extracts what you need. Read the file only with offset/limit. See README → Big-output discipline.