mk:chom

What This Skill Does

Chom analyzes external codebases, web apps, or described ideas to produce a Replication Spec -- a structured decision document that answers whether and how to replicate the feature in your project. It runs through a 6-phase workflow ending at a HARD GATE that requires human approval before any code is written. Chom does NOT write code; it produces a spec for downstream skills (mk:plan-creator, mk:cook) to implement.

When to Use

Triggers:

• "copy this from X", "replicate this feature", "clone how Y does Z", "port from", "build like X"
• "how does X do Y", "compare X vs our approach"

Anti-triggers:

• Packing the current project for export -- use mk:pack
• One-shot URL-to-markdown fetches -- use mk:web-to-markdown
• Exploring your own codebase architecture -- use mk:scout

Core Capabilities

• Multi-source input routing -- git repos (shallow clone + scout), web URLs (fetch + browse), local paths (direct read), freeform text (researcher agent)
• 6-phase analysis workflow with a non-bypassable HARD GATE at Phase 4
• 4 adaptation modes: --compare (side-by-side only), --copy (minimum changes), --improve (replace anti-patterns), --port (idiomatic rewrite)
• Structured challenge framework -- 7 questions scoring necessity, stack fit, data model, dependency cost, effort/value, blast radius, and maintenance burden
• Risk scoring -- 0-2 reds = low (proceed), 3-4 = medium (resolve first), 5+ = high (reject)
• Escalation for complex flows -- 3+ architectural layers trigger a handoff to mk:sequential-thinking before continuing

Arguments

Flag	Effect
--compare	Side-by-side analysis only (phases 1-3). No decision or handoff.
--copy	Bias toward minimum-adaptation transplant. Focuses on compatibility gaps.
--improve	Bias toward replacing anti-patterns during port.
--port	Bias toward idiomatic rewrite for your stack.
--lean	Skip researcher agent background gathering. HARD GATE still enforced.
--auto	Auto-approve non-HARD-GATE steps. HARD GATE still requires human approval.

No flag: full 6-phase workflow, emits Replication Spec without declaring adaptation depth.

Workflow

[1. Recon] -> [2. Map] -> [3. Analyze] -> [4. Challenge] == HARD GATE == [5. Decision] -> [6. Handoff]

1. Recon -- Route input by type (git clone/scout, web fetch, local read, researcher agent). Always read docs/project-context.md for your stack.
2. Map -- Build dependency matrix: source components mapped to local equivalents (EXISTS / NEW / CONFLICT).
3. Analyze -- Trace execution flow, data model, API contracts, UX patterns. Mode-specific focus for --copy/--improve/--port.
4. Challenge -- Load references/challenge-framework.md. Ask 7 questions. Score risk. Present decision matrix. Human approval required.
5. Decision -- Go/no-go. If go: formalize Replication Spec. If no-go: Rejection Report.
6. Handoff -- Output spec with risk score, red flag summary, and suggested next command.

Usage

/mk:chom https://github.com/vercel/ai --port
/mk:chom "copy the dark mode toggle from linear.app" --compare
/mk:chom ./path/to/local/repo --copy --lean

Example Prompt

/mk:chom https://github.com/shadcn-ui/ui --improve "We want their component composition pattern but adapted for MUI instead of Radix"

Common Use Cases

• Evaluating whether to replicate a competitor's feature before building
• Porting an open-source library's pattern into your stack idiomatically
• Side-by-side architectural comparison of two approaches
• Pre-build validation: does this feature fit our stack, or should we reject it?
• Understanding how an external system solves a problem before designing your own

Pro Tips

• Always pass a mode flag. No-flag runs work but don't bias analysis toward your intent. If you know you want to port, say --port.
• Read docs/project-context.md first. Without it, chom gives generic advice that won't fit your stack.
• HARD GATE is your safety net. The challenge phase catches stack mismatches, hidden dependencies, and maintenance burdens that look obvious in hindsight.
• --compare is cheap. When unsure about replication, run --compare first to see the head-to-head before committing to a full spec.
• Chom never chains into other skills. Handoff text tells you what to run next (mk:plan-creator, mk:sequential-thinking). You invoke it.

Canonical source: .claude/skills/chom/SKILL.md