Minimal modular code
Skill justinramos101/agent-skill-kit/skills/.experimental/minimal-modular-code
Keep code minimal and modular: trim over-engineered or slop-heavy AI-generated code, keep a change small by reusing what exists and subtracting before adding, decide whether deleting code (guards, tests, validation, error paths) is safe, place module boundaries and dependency direction, and partition a repo so parallel agents don't collide. Triggers: 'over-engineered', 'reduce slop', 'too much abstraction', 'should I delete this guard/test', 'duplication', 'module boundaries', 'coupling', 'split work across agents'. Do NOT use for wiring enforcement hooks/CI gates/AGENTS.md, developer-facing API/SDK/CLI surface design, product UX, or prose editing.From its SKILL.md
npx -y skills add justinramos101/agent-skill-kit --skill minimal-modular-codeAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
3.3 KB, 555 tokens by cl100k_base, as published. Nobody here has run it
Minimal Modular Code
Keep agent-written code minimal and modular. Minimal is not fewest lines: it is behavior behind a small interface, reused before rebuilt, legible to a finite-context reader. Steer by named heuristics:
- Reuse before write — grep the repo and deps for an existing helper before adding one.
- Subtract first — ask "what can be removed?"; the subtractive option never surfaces on its own.
- Rule of three — duplication is cheaper than the wrong abstraction; back a bad abstraction out rather than parameterize it.
- Deep modules — small interface over substantial behavior; single-caller indirection and pass-through layers are negative-value structure.
- Information hiding — put the seam where it hides the likely-to-change decision; dependencies point toward stable code.
- Concise, not terse — fewer semantic elements, never code golf.
- Gates, not prose — a load-bearing invariant needs a failing check, not a sentence.
Deletion needs proof. Never delete guards, tests, validation, error paths, or boundaries to make a diff smaller unless the prompt supplies proof they are obsolete — from blame, tests, the contract, or an upstream invariant. "Not in scope", downstream behavior, or a promised later PR is not proof. When proof is missing, keep the code and say why.
Workflow
- Restate the ask: smallest safe change, slop/coupling audit, or boundary design. Infer it from the request.
- Read the touched code and its callers; search for existing helpers before proposing new ones. Compute the blast radius: what does the change touch beyond the stated scope?
- Load the one reference the task needs:
references/minimalism.mdfor amount-of-code and readability questions (slop, duplication, over-engineering, deletion, cleverness);references/boundaries.mdfor seam, coupling, parallel-work, and gate-coverage questions. - Propose the minimal change or the findings. Preserve unrelated names, control flow, and interfaces. Snippets include imports for newly referenced types; with partial context keep multi-file details conceptual rather than invented.
- When reviewing, severity is one line: blocking (violated invariant, wrong-direction dependency, deleted safety behavior) vs advisory (style, deferred cleanup).
Stop conditions
- Stop when the requested behavior is achieved and nothing more can be removed with proof. Do not expand a fix into a rewrite; opportunistic cleanup is out unless the fix requires it.
- Run to completion without check-ins: state assumptions inline and proceed. Ask only on a genuine blocker — a missing file or a contradictory requirement.
What ships with it: 9 files
48.0 KB alongside SKILL.md, 1 of them executable
evals/
- activation-cases.md3.3 KB
- behavior-cases.json10.3 KB
- behavior-eval.md4.7 KB
- run-static-checks.shruns1013 B
- trigger-evals.json4.1 KB
references/
- boundaries.md4.2 KB
- minimalism.md3.7 KB
- README.md2.0 KB
- skill.json14.7 KB