Precedent Finder

Overview

Recommend 5-8 architectural precedents based on a designer's natural-language brief, with each entry annotated for the strategy (not feature) the designer should learn from. Cache the result so future searches on the same topic short-circuit and the team builds a project-specific precedent library.

When to use

Trigger on:

Concept/atmosphere descriptions ("a museum embedded into a hillside, warm concrete not cold").
Program + site briefs ("30,000 sf riverside cultural center, mass timber").
Explicit asks: "precedents for", "references", "similar projects", "comparable buildings".

Do NOT trigger for general design guidance ("how should I approach this site"). This skill returns examples, not method.

Workflow

Step 1 — Bootstrap if needed

Check docs/precedents/index.md exists. If not:

python scripts/init_precedents_workspace.py

Step 2 — Parse the brief

Decompose into searchable axes:

Program (museum / housing / cultural / mixed-use / etc.)
Scale (rough area, units, budget tier)
Site condition (urban / coastal / mountain / hillside / industrial)
Atmosphere keywords (warm, brutal, ceremonial, domestic)
Material / form keywords (concrete, mass timber, cantilever, courtyard)

Optionally read docs/project.md for additional project context.

Step 3 — Check the precedent canon

Before generating from memory, load references/canon.md for a curated list of well-documented precedents organized by program and atmosphere. Use it as the default pool to draw from — augment with additional projects only when none in the canon match.

Step 4 — Recommend 5-8 precedents

Deliberately span:

Different eras (don't recommend only contemporary)
Different geographies (avoid clustering all from one region)
Different scales (include at least one smaller and one larger than user's project)
At least one canonical pick AND at least one unexpected pick

Per precedent:

Architect (firm) / completion year / location / approximate area
Three core design strategies (not features)
The match: which axis of the brief this addresses
Source link (ArchDaily, Dezeen, architect's official site, monograph)

Step 5 — Write the precedent file

Read assets/precedent-template.md. Fill placeholders. Filename:

docs/precedents/<topic-slug>.md

Slug rule: lowercase, [^a-z0-9-] removed, max 40 chars, derived from the brief's dominant keywords (hillside-museum-concrete, riverside-cultural-mass-timber). If duplicate, suffix -<n>.

Step 6 — Update the precedent index

python scripts/append_to_index.py \
  --date 2026-05-06 \
  --slug hillside-museum-concrete \
  --program museum \
  --keywords "hillside,concrete,warm" \
  --count 6

Step 7 — Report

Tell the user:

Path to the precedent file
The 1-2 strongest matches with one-line reason
Optionally: "Want me to expand on any of these or pull more in a different direction?"

Rules

A "strategy" is a design move, not a feature. "Embedded the mass into the hillside so the roof becomes the site" is a strategy. "Has a green roof" is a feature.
Cite the strongest single match per project — do not list every possible reason.
If the brief is too vague ("modern museum"), ask one clarifying question (atmosphere or site) before producing the list.
Default to North American and international canon. Include local references where the brief specifies a region.
Do not recommend the user's own work or unbuilt competition entries unless asked.

Anti-patterns

Padding the list with weak matches to hit a number. 5 strong is better than 10 mixed.
Summarizing Wikipedia. Provide design-strategy reasoning a designer can use immediately.
Inventing projects. If unsure of existence, year, or architect, omit the entry.
Skipping the canon and going straight to memory — the canon is there to anchor.

Resources

scripts/init_precedents_workspace.py — bootstrap docs/precedents/ and the index
scripts/append_to_index.py — append a row to docs/precedents/index.md
references/canon.md — curated list of well-documented precedents by program/region/era
assets/precedent-template.md — precedent file template

What this does. Reads a brief in the words you would use to describe it and returns five to eight built precedents, each with the design strategy worth taking from it and the part of your brief it answers. The list is cached in docs/precedents/ so the next search on the same topic starts from it.

What this does not do.

It does not invent a project. Name, architect, year, and location each have to be verifiable to a primary source or the entry is left out. references/canon.md is the pool it draws from first, and every entry there was checked against a published record.
It does not search the web. What it knows about a project can be out of date, which is why every entry carries a source link for you to open before citing it.
It does not analyse a precedent in depth. It names the strategy; the drawings and the monograph are still yours to read.
It does not tell you how to design. It returns examples, not method.
It does not offer unbuilt competition entries or your own past work unless you ask for them.

What you need before starting. The program, the atmosphere or site condition you are after, and a scale if you have one. A brief as thin as "modern museum" gets one clarifying question before the list is built.

Precedent Finder

Overview

When to use

Trigger on:

Concept/atmosphere descriptions ("a museum embedded into a hillside, warm concrete not cold").
Program + site briefs ("30,000 sf riverside cultural center, mass timber").
Explicit asks: "precedents for", "references", "similar projects", "comparable buildings".

Do NOT trigger for general design guidance ("how should I approach this site"). This skill returns examples, not method.

Workflow

Step 1 — Bootstrap if needed

Check docs/precedents/index.md exists. If not:

python scripts/init_precedents_workspace.py

Step 2 — Parse the brief

Decompose into searchable axes:

Program (museum / housing / cultural / mixed-use / etc.)
Scale (rough area, units, budget tier)
Site condition (urban / coastal / mountain / hillside / industrial)
Atmosphere keywords (warm, brutal, ceremonial, domestic)
Material / form keywords (concrete, mass timber, cantilever, courtyard)

Optionally read docs/project.md for additional project context.

Step 3 — Check the precedent canon

Step 4 — Recommend 5-8 precedents

Deliberately span:

Different eras (don't recommend only contemporary)
Different geographies (avoid clustering all from one region)
Different scales (include at least one smaller and one larger than user's project)
At least one canonical pick AND at least one unexpected pick

Per precedent:

Architect (firm) / completion year / location / approximate area
Three core design strategies (not features)
The match: which axis of the brief this addresses
Source link (ArchDaily, Dezeen, architect's official site, monograph)

Step 5 — Write the precedent file

Read assets/precedent-template.md. Fill placeholders. Filename:

docs/precedents/<topic-slug>.md

Slug rule: lowercase, [^a-z0-9-] removed, max 40 chars, derived from the brief's dominant keywords (hillside-museum-concrete, riverside-cultural-mass-timber). If duplicate, suffix -<n>.

Step 6 — Update the precedent index

python scripts/append_to_index.py \
  --date 2026-05-06 \
  --slug hillside-museum-concrete \
  --program museum \
  --keywords "hillside,concrete,warm" \
  --count 6

Step 7 — Report

Tell the user:

Path to the precedent file
The 1-2 strongest matches with one-line reason
Optionally: "Want me to expand on any of these or pull more in a different direction?"

Rules

A "strategy" is a design move, not a feature. "Embedded the mass into the hillside so the roof becomes the site" is a strategy. "Has a green roof" is a feature.
Cite the strongest single match per project — do not list every possible reason.
If the brief is too vague ("modern museum"), ask one clarifying question (atmosphere or site) before producing the list.
Default to North American and international canon. Include local references where the brief specifies a region.
Do not recommend the user's own work or unbuilt competition entries unless asked.

Anti-patterns

Padding the list with weak matches to hit a number. 5 strong is better than 10 mixed.
Summarizing Wikipedia. Provide design-strategy reasoning a designer can use immediately.
Inventing projects. If unsure of existence, year, or architect, omit the entry.
Skipping the canon and going straight to memory — the canon is there to anchor.

Resources

scripts/init_precedents_workspace.py — bootstrap docs/precedents/ and the index
scripts/append_to_index.py — append a row to docs/precedents/index.md
references/canon.md — curated list of well-documented precedents by program/region/era
assets/precedent-template.md — precedent file template

What this does not do.

It does not invent a project. Name, architect, year, and location each have to be verifiable to a primary source or the entry is left out. references/canon.md is the pool it draws from first, and every entry there was checked against a published record.
It does not search the web. What it knows about a project can be out of date, which is why every entry carries a source link for you to open before citing it.
It does not analyse a precedent in depth. It names the strategy; the drawings and the monograph are still yours to read.
It does not tell you how to design. It returns examples, not method.
It does not offer unbuilt competition entries or your own past work unless you ask for them.

Precedent Finder.

Precedent Finder

Overview

When to use

Workflow

Step 1 — Bootstrap if needed

Step 2 — Parse the brief

Step 3 — Check the precedent canon

Step 4 — Recommend 5-8 precedents

Step 5 — Write the precedent file

Step 6 — Update the precedent index

Step 7 — Report

Rules

Anti-patterns

Resources

Precedent Finder.

Precedent Finder

Overview

When to use

Workflow

Step 1 — Bootstrap if needed

Step 2 — Parse the brief

Step 3 — Check the precedent canon

Step 4 — Recommend 5-8 precedents

Step 5 — Write the precedent file

Step 6 — Update the precedent index

Step 7 — Report

Rules

Anti-patterns

Resources