field-guide-builder
Skill1 starsChanged 56 days agoMITView on GitHub
Use when building or extending a single-file, source-grounded, visually distinctive teaching guide on a technical topic, the kind with a concept hierarchy of worked examples and do/avoid practice, inline citations to primary sources, an editorial design system, an in-app router, and an optional interactive companion. Triggers on requests like "build a guide", "make a teaching guide / explainer / field guide / learning doc", or polishing and extending one.
In this repository64 other files it gives its agents.
- accioSkill
- anthropic-doc-validatorSkill
- archifySkill
- cavecrewSkill
- caveman-compressSkill
- caveman-exploreSkill
- caveman-helpSkill
- cavemanSkill
- codebase-designSkill
- conference-talk-deckSkill
- config-gcSkill
- context-budgetSkill
- domain-modelingSkill
- excalidrawSkill
- explain-yogi-like-he-is-5Skill
- gateguardSkill
- geo-contentSkill
- goal-creatorSkill
- grillingSkill
- grill-meSkill
- grill-with-docsSkill
- hindsightSkill
- implementSkill
- improve-codebase-architectureSkill
- loop-design-checkSkill
- mental-modelsSkill
- migrationSkill
- modern-pythonSkill
- no-ai-slopSkill
- onboard-lightSkill
- onboard-repoSkill
- ponytail-auditSkill
- ponytail-reviewSkill
- ponytailSkill
- primitive-classifierSkill
- prompt-fixerSkill
- prompt-optimizerSkill
- prototypeSkill
- qa-deckSkill
- read-wikiSkill
- researchSkill
- resolving-merge-conflictsSkill
- safe-refactorSkill
- scope-creep-checkSkill
- sdd-workshop-walkthroughSkill
- setup-matt-pocock-skillsSkill
- skill-complySkill
- skill-scoutSkill
- smellsSkill
- spec-interviewSkill
- surgical-patchSkill
- teachSkill
- the-llm-councilSkill
- to-questionnaireSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- wait-whatSkill
- watchSkill
- wayfinderSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 32
Build the structure: Add the nav, the hash router, and one concept card plus its two deep-dive pages as a template …
- Very specific: line 35
QA and ship: Run assets/qa.py. Verify routing, search, every citation resolves, every anchor resolves, and motion respects reduced-motion.
- Distinctive: line 39
No em dashes anywhere (use commas, colons, parentheses, or restructure).
- Very specific: line 40
Every [N] citation resolves to a real reference.
- Distinctive: line 41
No claim without a source.
- Distinctive: line 45
Verify in a real browser (local server + a headless check), not by eye alone.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Field guide builder
Build the kind of guide we build: one self-contained HTML file that teaches a technical topic to a high standard, with distinctive design, rigorous sourcing, and one consistent teaching hierarchy. Optionally ship an interactive companion (a step-through "session") next to it. The canonical instance this skill was distilled from is a prompting guide, but the recipe is topic-agnostic.
This file is the recipe. The details live in reference files; load only the one the current phase needs (that is the same progressive-disclosure principle these guides teach):
references/design-system.md, palette, fonts, the masthead seal, component CSS, atmosphere, motion.references/structure-and-router.md, the concept hierarchy, deep-dive pages, nav, hash router, search, citations.references/sourcing-and-qa.md, primary-source rigor, the parallel-agent hallucination audit, the QA checklist.references/interactive-companion.md, the conversation-starters / step-through session pattern.assets/qa.py, run against the finished file to check em dashes, citation integrity, and links.
The four pillars (what makes it "this kind of guide")
- Single file. One
.html, no build step, inline<style>and<script>, web-font links only. It opens anywhere and is trivially shareable. - Source-grounded. Every concrete claim (number, date, feature, behavior, quote) traces to a primary source: official docs, the vendor's engineering blog, a named conference talk, or the creator's own words. Secondary blogs may support a claim but never anchor it. No invented stats or features. Cite inline with
[N]linking to a references list. - One teaching hierarchy, everywhere. Each concept is a card: a one-line definition, then Example (2 good + 1 bad, worked) and In Practice (Do / Avoid). Heavy concepts also get full deep-dive pages for each half, reached through an in-app router. Same shape for every concept, no exceptions.
- Editorial, not generic. A committed visual identity (a distinctive display font paired with a refined body font, one dominant color with sparing accents, real atmosphere, and a signature mark), never framework defaults or "AI slop". See the design system.
Build workflow
Work in phases. Do not try to do it all at once.
- Scope and source-gather first. List the concepts. For each, find the primary source and pull the exact facts (quotes verbatim, numbers exact, dates exact). Keep a source map as you go. If a claim cannot be sourced, cut it or mark it plainly as an illustrative example.
- Lay down the design system. Drop in the tokens, fonts, masthead, and base components from
design-system.md. Make the empty shell look finished before pouring in content. - Build the structure. Add the nav, the hash router, and one concept card plus its two deep-dive pages as a template (
structure-and-router.md). Verify routing in a browser before scaling. - Fill content, concept by concept. Reuse the card + deep-dive template per concept. Color-code by category. Keep every Example at 2 good / 1 bad, every In Practice with concrete Do/Avoid items and a routing table.
- Audit for hallucinations. Extract every claim and verify it against the live primary source. For a large guide, fan out parallel research agents over claim-clusters (one per section group) to flag anything unverifiable or embellished, then fix. See
sourcing-and-qa.md. - QA and ship. Run
assets/qa.py. Verify routing, search, every citation resolves, every anchor resolves, and motion respects reduced-motion. Optionally build the interactive companion (interactive-companion.md).
Non-negotiables
- No em dashes anywhere (use commas, colons, parentheses, or restructure). En dashes are fine for ranges. QA before declaring done.
- Every
[N]citation resolves to a real reference; every internal#anchorresolves both ways. - No claim without a source. Verbatim quotes are actually verbatim. Numbers and dates are exact, not "about right".
- Distinguish a factual claim (must be true) from an illustrative example (a made-up snippet that demonstrates a real concept is fine; an invented statistic is not).
- One consistent teaching hierarchy across all concepts.
- Animations gated behind
prefers-reduced-motion, with at most one tiny deliberate exception you can justify. - Verify in a real browser (local server + a headless check), not by eye alone.
House rules
Honor the user's stated writing preferences (for this user: never em dashes; en dashes only for numeric or date ranges). When extending an existing guide, read its tokens and conventions first and match them exactly rather than introducing a second style.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

