Option Compare
Build a side-by-side comparison of design options for a decision
Once installed, Claude loads it on its own when your conversation matches. You can also call it directly with /option-compare.
Install just this one
npx archtmpl@latest --skill option-compare --globalFirst time? The whole install, step by step
- Open Claude Code — the terminal version or the desktop app, either one.
- In a terminal, paste the line above and press Enter. In the app, paste it into the chat and ask Claude to run it.
- Restart Claude Code. That's the whole install.
Set up plugins for me: run `claude plugin marketplace add https://archaiflow.com/plugins/marketplace.json` and then `claude plugin install option-compare@archaiflow`Paste into the Code tab (not Chat or Cowork) and approve when Claude asks. The third-party marketplace it mentions is this site. Windows may ask to install Git once.
What this skill does
Option Compare
Every early design phase ends the same way: two or three schemes, one meeting, and a decision that will be relitigated forever if nobody wrote down what was traded for what. The comparison matrix is how the decision gets made on the table instead of in the hallway. This builds it, keeps every cell honest about where it came from, and leaves the choosing to the people whose choice it is.
Workflow
Step 1. Bootstrap if needed
Check whether docs/design/options.md exists; create it if not, one
section per comparison round, dated. If the project has no docs/
folder at all, confirm before creating one.
Step 2. Frame the decision
Ask, in one message:
- What is being decided, and by whom, at what meeting?
- Which options are on the table? Names and a one-line description each, as you would say them to the client.
- Which criteria should the decision turn on? I can suggest common blanks, but this list is yours.
- Where do values come from: typed, a space-program file, a cost source?
Step 3. Fill the matrix
One row per criterion, one column per option. Every cell records its
source: you, a named file, a cited benchmark. Where a space-program
file exists, offer to pull each option's areas from it and cite the
file. Quantitative rows keep units as supplied; qualitative rows keep
your wording, quoted, not paraphrased upward.
| Criterion | Option A | Option B | Source |
|-----------------|------------------|------------------|-----------|
| <as chosen> | <as supplied> | <as supplied> | <each> |
| <...> | not provided | <as supplied> | |
Step 4. Score, only if weighted
If you supply weights, normalize them in the open, multiply, sum, and show every step. No weights, no scores; a matrix quietly ranked by an unstated weighting is the most common way these documents lie.
Step 5. Write the tradeoff summary
One short paragraph per option, parallel in structure: what choosing it
buys, what it costs, stated from the matrix rows and nothing else. Then
the open questions: every not provided cell, listed as what would have
to be known for the comparison to be complete.
Step 6. Record the decision
When the decision lands, record it under the round: which option, who
decided, the date, and the why in their words. Offer to log it to
project-tracker as a decision entry too; that log is where "why is
the building L-shaped" gets answered in a year.
Step 7. Export, if the meeting needs paper
Offer aec-doc-gen to render the round to Word or PDF for the handout.
Rules
- Every cell carries a source. No source, the cell reads
not provided, and the summary counts them. - Parallel wording across columns. "Efficient core" for A and "awkward core" for B is a decision wearing a matrix costume; state the measurable difference or quote the stated judgment with its author.
- No scores without supplied weights, and no recommendation either way.
- Rounds append. A revised comparison is a new dated section; the matrix the client saw in March stays as it was.
- Quote judgments, with their author. "Client prefers the porch (meeting 3/14)" is data; "warmer entry experience" from nowhere is not.
Anti-patterns
- Filling an obvious cell (the bigger scheme "probably costs more") to complete the picture.
- Wording one column warmly and its neighbor clinically.
- Inventing a criterion because comparisons usually have one.
- Averaging two stakeholders' conflicting judgments into one cell.
- Deleting a superseded round instead of dating a new one.
Resources
None. This skill is one file. The matrix is written directly to
docs/design/options.md.
What it does not check
What this does. Collects the schemes, asks which criteria matter for this decision, fills the matrix from what you supply or from project files you point it at, marks every cell with its source, writes the tradeoff summary each option's column implies, and records the eventual decision with its date and its why.
What this does not do.
- It does not fill cells on its own. An area you did not supply, a
cost nobody stated, a "feels more open": none of it enters the matrix.
Empty cells print as
not provided, and a column with three of them reads as the less-studied option it is. - It does not pick the winner. The summary names what each option gives up and gains, in parallel wording. Ranking happens only under weights you supply, with the arithmetic shown, and even then it prints as "your weights rank B first", not as a recommendation.
- It does not design or improve the options, and it does not invent a
compromise scheme D.
design-criticcritiques a scheme; this compares what exists. - It does not benchmark. A cost cell wants your number or a cited
benchmark you accepted (
lookup-costis the tool for finding one). - Criteria are yours. It proposes the blanks a decision usually needs (program fit, area, cost basis, schedule, code posture) as questions, never as filled rows.
What you need before starting. Two or more schemes that actually exist, the criteria the decision turns on, and values for the cells you want compared, with sources.
Files it puts on your disk
.claude/skills/option-compare/1 file · 6.2 KBSKILL.md6.2 KB