Precedent Critique

Take the projects you admire, work out from their own material what they actually do, and read your scheme through that. The lenses come from what you hand over, never from a remembered idea of a firm's style.

When to use

Use design-critic for a general second opinion against its fixed six axes. Use this skill when you have specific work you want your scheme held next to, and you can put that work's material on the table.

precedent-finder is the natural first step when you have the sensibility but not the list: it names candidate projects with source links, and you come back here with the material you gathered from them.

Workflow

Step 1. Collect both sides

Ask for anything missing, in one message:

1. The scheme: concept text, plans, sections, images, in whatever state it is in, plus the site and program in a sentence each.
2. The references: two to four projects, each with material. Paste the text, attach the images, or name the files.
3. Optionally, what drew you to these references, in your own words. This is checked against what the material shows, not substituted for it.

If a reference arrives as a name only, say plainly that a name is not material and ask for some, or point to precedent-finder to build the list first. Do not fill the gap from memory, and do not proceed with the references that have material while silently dropping the one that does not.

Step 2. Extract the lenses

Read the reference material and pull out the three to five moves that recur across it: how the buildings meet the ground, what the section does that the plan does not, where material restraint sits, what gets spent and what gets suppressed. Each lens must cite the supplied material that shows it, by reference name and by the specific passage, drawing, or image. A lens that cannot point at something in the material is dropped, not kept on conviction.

A move that appears in one reference only is noted as that project's move, not elevated into a lens. A lens is what the set shares.

Step 3. Confirm the lenses

Show the lenses before critiquing, each with its citations, and ask two things: are these the qualities you were after, and which one matters most. You may strike a lens, add one by pointing at material that shows it, or re-rank. Do not skip this step; a critique built on a misread of the references is worse than no critique.

Step 4. Read the scheme through each lens

For each confirmed lens, one of three findings, and only three:

• Aligned, with the place in your scheme where the lens shows up, named specifically.
• Diverges, with what the scheme does instead, and the question stated plainly: is this chosen, or did it drift? A chosen divergence is a position; a drifted one is the finding.
• Not visible in what you shared. Say what material would settle it. Do not guess this state into either of the other two.

Step 5. Report

Close with the two or three findings worth acting on, each tied to its lens and to something visible in the work, and the single next study that would move the scheme most. No scores, no ranking against the references, no verdict on taste.

Then ask whether to save the critique, and where. A sensible default is docs/reviews/<YYYY-MM-DD>-precedent-<slug>.md, written directly, alongside any design-critic reviews already there. There is no template and no script.

Rules

• Never characterize a reference from memory. The material on the table is the whole of what the skill knows about the project. If it is not enough, say so and ask, rather than quietly topping it up.
• Every lens carries its citations into the output, and every finding names both the lens and the place in the scheme it points at. "This feels un-reference-like" with no anchor is not a finding.
• Divergence is a question, not a fault. The report asks whether it was chosen; it never subtracts points for it.
• Three finding states, no third guessed into the first two. What the shared material cannot show is reported as not visible.
• Keep it bounded: three to five lenses, two or three findings worth acting on. Ten lenses means the references were read too thin to rank what recurs.

Anti-patterns

• Running with "critique this like BIG" and no material. That produces a style caricature with a real firm's name attached, which is exactly what this skill exists to refuse.
• Topping up a thin excerpt with what the model knows about the project. The supplied material is the boundary, famous or not.
• Grading the scheme against the references: "they do X, you do not, minus one". The references are a way of seeing, not a rubric.
• Elevating one reference's one-off move into a lens for the set.
• Critiquing material the user only alluded to, or inferring the site and program instead of asking once.
• Suggesting the scheme become one of the references. The critique points to the next move in this project, not to a copy of another.

Resources

None. This skill is one file. Output is written directly at the path you choose.