Fuck Over-Engineering

Audit the scope for over-engineering and report what to cut, ranked biggest cut first. The best outcome is a shorter codebase with the same behavior. The argument is the scope; without one, the whole tree. Read what the scope holds before judging it, and fan out subagents when it will not fit in context.

Tags

• delete: dead code, unused flexibility, speculative feature. Nothing replaces it.
• reuse: hand-written code that a helper already in this codebase, or an installed dependency, already provides. Name the helper or function.
• stdlib: hand-rolled thing the language's standard library ships. Name the function.
• native: dependency or code doing what the platform already does. Name the feature.
• yagni: abstraction with one implementation, config nobody sets, layer with one caller.
• shrink: same logic, fewer lines. Show the shorter form.

Hunt

Dependencies the stdlib or platform already covers, single-implementation interfaces, factories with one product, wrappers that only delegate, files exporting one thing, dead flags and config, hand-rolled stdlib, near-duplicates of an existing helper, special-case branches bolted onto shared paths, feature logic living in shared modules.

Evidence

Every finding cites the evidence that makes it a cut: caller count, implementation count, or the stdlib or platform feature and the version that ships it. Count callers through dynamic dispatch, entry points, hooks, and tests, not grep alone. A published package's public export stays at zero in-repo callers. Verify stdlib: and native: claims against current docs (the find-docs skill when present) before asserting them.

A finding removes lines or concepts; moving them between files is a refactor, not a cut. Fewer, bigger cuts beat a long list: a shrink: that saves two lines earns a slot only when it also removes a concept.

Output

Use AskUserQuestion in Claude Code or request_user_input in Codex when available. Use plain text when neither tool is available.

One line per finding, numbered, ranked biggest cut first:

N. <tag> <what to cut>. <replacement>. <evidence>. [path:line]

delete: omits the replacement sentence. End with net: -<N> lines, -<M> deps possible. Nothing to cut: Lean already. Ship.

• 1. yagni: AbstractRepository with one implementation. Inline it. 1 subclass, 3 call sites typed to the base. [repo.py:88]
• 2. reuse: hand-rolled slugify. python-slugify is installed, slugify(). 1 caller. [utils/text.py:10-31]
• 3. delete: retry wrapper around an idempotent local call. 0 callers outside its own test. [src/net.py:52-71]

Then offer the findings in ranked batches, collect the user's picks, and apply them.

Boundaries

Over-engineering and complexity only: correctness bugs, security holes, and performance belong to a normal review pass. Trust-boundary validation, data-loss handling, security, and accessibility are never cuts. A single smoke test or assert-based self-check is the minimum, not bloat. Tests enter the report only as a side effect of a deleted target.