Overkill — Architecture Exploration Beyond the Pragmatic Answer
Proposes 3–6 advanced alternatives past your pragmatic solution — advanced data structures, distributed algorithms, niche frameworks, frontier tooling — each scored on a calibrated complexity scale.
A pure prompt-based skill: no scripts, no side effects, no sensitive data handling; --current web access is explicit opt-in, default mode is offline and deterministic, data flow transparent. Deducted because behavior in --current is enforced only by prompt instructions rather than mechanisms, and there is no third-party execution evidence, so full marks not awarded.
Instructions are self-consistent and well-structured, with example output and explicit graceful degradation when web access is unavailable. Deducted for: no tests, no execution evidence, 🔥 cross-invocation calibration depends on model discipline and may drift; failure-feedback quality unverifiable statically — capped at the static ceiling of 10.
Trigger conditions ('overkill', '/overkill', flags) are precise, non-fit boundaries are explicitly declared, and a clarifying-question rule exists. Deducted for: English only, no Chinese support; --current depends on reachability of overseas services from mainland China (mitigated by graceful fallback), so not full marks.
Clear installation paths (marketplace and manual), complete MIT license, well-layered documentation. Deducted for: no version number, no changelog, no declared maintenance responsibility or update path, publisher identity unverified — below the 15 anchor.
Concrete output template, complete worked example, clear comparative value (expanding the design space, not default recommendation), proportionate cost. Deducted because static review cannot verify real output quality or the claimed cross-invocation calibration of the complexity scale — capped at 7.
Example includes auditable primary sources (arXiv paper, official docs). Deducted for: no CI, no test suite, no third-party execution evidence; static sources only with thin coverage — capped at 5.
- Static source review only; actual output quality and cross-invocation calibration of the complexity scale are unverified.
- English-only output; no Chinese-language support.
- --current mode depends on overseas web/search reachability and may be unavailable from mainland China (the skill declares a graceful offline fallback).
- Publisher unverified; no versioning or maintenance commitment, so future upkeep is unknown.
- The skill deliberately proposes maximalist options and must not be treated as a default production recommendation.
What does this skill do, and when should you use it?
Overkill is a Claude Code skill for when you already have a working, pragmatic solution and want to see what lies beyond it. It restates your baseline, then proposes 3–6 advanced alternatives drawn from distinct categories — data structures, algorithms, frameworks, design patterns, and tooling — ordered from well-known to obscure and scored on a calibrated 🔥 1–10 complexity scale. Each option comes with durable learning links, the concrete skills the path develops, and the future-scale scenario where it becomes the right choice. It is explicitly not a pragmatic recommendation engine; it expands your map of what is possible.
The skill reads the problem currently under discussion and restates the baseline in one sentence. It then generates 3–6 cross-category alternatives, each with: actual mechanics, how it exceeds the baseline, skills developed, 1–3 durable links (papers and primary docs preferred), and a concrete payoff scenario. Results are presented in a comparison table — learner-focused by default, or operator-focused under --advanced (time to first commit, ops burden, hiring difficulty). The optional --current flag uses web search to verify project health and refresh references; --max restricts output to complexity 🔥 7+ options only.
- An engineer implementing a feature wants to survey the full design space before committing to the simple option
- A staff or principal engineer preparing a tech-design doc needs a survey of advanced alternatives to cite and rule out
- A team doing technical due diligence on a vendor's claim ("they use X") wants to know where X actually sits on the complexity spectrum
- An engineer choosing a learning project wants a study path anchored to a real problem rather than an abstract topic
- A platform lead in --advanced mode evaluating the real adoption cost of a frontier technology — ops burden, hiring, time to first commit
What are this skill's strengths and limitations?
- Complexity scores are anchored against an explicit cross-response rubric (🔥 1–10), preventing score drift
- Enforces category variety — never six frameworks in a row
- --current mode actively verifies recommended projects are maintained and degrades gracefully offline with an explicit notice
- Clearly scoped: states outright it should not be used as a default production recommendation, avoiding misuse
- By design it generates deliberate overkill — adopting its suggestions as default architecture is misuse
- --current depends on web search availability; offline it falls back to training-cutoff references and adds latency
- No visible test suite or cross-platform verification in the repository; empty Topics field and limited third-party adoption evidence
- The complexity scale is a model-calibrated self-rating, not an externally validated metric
How do you install this skill?
Option 1 (recommended, via plugin marketplace): inside any Claude Code session run /plugin marketplace add santiago-vargas-de-kruijf/claude-overkill, then /plugin install overkill@overkill-marketplace. Option 2 (manual): clone the repository and copy the plugins/overkill/skills/overkill folder to ~/.claude/skills/ (personal, available in every project) or to .claude/skills/ in a project (shared via git). The README documents no Windows-specific steps or installs for other platforms.
How do you use this skill?
In a Claude Code conversation about a concrete problem, trigger with overkill, overkill this, or /overkill. With no flags, it first shows a multi-select mode picker; passing any flag skips it: overkill --max (only 🔥 7+ frontier options), overkill --advanced (operator-focused comparison table), overkill --current (web search to verify references and project health, adds latency). Flags combine freely. Output includes ranked alternatives with complexity scores, a comparison table, and the concrete scale or condition under which each option earns its cost.