agentleFS
diegopachecodiegopacheco/agent-skills/html-research/SKILL.md

html-research

Skill0 starsChanged 3 months agoUnlicenseView on GitHub

Research a topic on the web and render one self-contained light-theme HTML report with searchable cards, clickable detail views, handwritten diagrams, file paths or GitHub links, and clickable sources. Use when the user runs /html-research or asks for a research report, a deep dive, a write-up, or how big tech solves a problem.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 4 are distinctive, and 9 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 9

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When the topic points at a local project, read the real files.

    line 15
  2. Very specific:

    Every report is saved to /Users/diegopacheco/git/diegopacheco/html-research/.

    line 19
  3. Distinctive:

    Run web research. Do not answer from memory.

    line 26
  4. Distinctive:

    If the topic is an engineering matter (rate limiting, auth, caching, sharding, cloud security, observability, deploys, feature flags, …

    line 27
  5. Very specific:

    If the topic touches a local repo, collect the real absolute paths.

    line 32
  6. Very specific:

    Run python3 scripts/check_report.py <report.html> and fix everything it reports.

    line 36
  7. Very specific:

    No CDN, no external stylesheet, no external script, no web font, no remote image.

    line 43
  8. Very specific:

    Only <a href="https://..."> links to sources are allowed to point outside the file.

    line 44
  9. Very specific:

    Every factual claim carries a clickable source.

    line 50
  10. Distinctive:

    Never fabricate a URL. If you did not open it, do not cite it.

    line 53
  11. Very specific:

    Add bullets, files, and links where they apply.

    line 70
  12. Very specific:

    Every report MUST have a What The Industry Does group.

    line 93
  13. Very specific:

    Use handwritten-style inline SVG.

    line 107

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

HTML Research

Turn a topic into one standalone HTML report backed by real web research.

Inputs

The topic is $ARGUMENTS. If it is empty, ask the user what to research and stop.

The topic can be a subject (rate limiting, auth, cloud security), a codebase question, or a mix. When the topic points at a local project, read the real files. Never invent paths.

Output Location

Every report is saved to /Users/diegopacheco/git/diegopacheco/html-research/.

Filename: {topic-slug}-{MM-yyyy}.html, lowercase, hyphenated.

Workflow

  1. Restate the topic in one line and name the angle you will research.
  2. Run web research. Do not answer from memory. Use several searches and fetch the strongest pages.
  3. If the topic is an engineering matter (rate limiting, auth, caching, sharding, cloud security, observability, deploys, feature flags, secrets, queues, and similar), you MUST research how big companies actually do it. Cover at least four of: Google, Amazon/AWS, Meta, Netflix, Stripe, Microsoft/Azure, GitHub, LinkedIn, Cloudflare, Shopify, Uber, or a relevant AI lab (Anthropic, OpenAI, DeepMind). Prefer engineering blogs, official docs, RFCs, papers, and talks.
  4. If the topic touches a local repo, collect the real absolute paths. If the file lives in a public GitHub repo, also collect its https://github.com/{org}/{repo}/blob/{ref}/{path} URL.
  5. Copy templates/report-template.html to the output path and fill the REPORT object.
  6. Draw diagrams only where they explain something a paragraph cannot.
  7. Run python3 scripts/check_report.py <report.html> and fix everything it reports.
  8. Run bash scripts/open_report.sh <report.html>.
  9. Print the full absolute path of the report as the last line of your answer.

Hard Rules

  • One file. No companion CSS, JS, JSON, images, or fonts. Everything inline.
  • No CDN, no external stylesheet, no external script, no web font, no remote image. Only <a href="https://..."> links to sources are allowed to point outside the file.
  • Light theme only. Never a dark palette, never a prefers-color-scheme: dark block, never a theme toggle.
  • Embedded favicon via a data:image/svg+xml <link rel="icon">. Keep it under 400 bytes.
  • Everything searchable through the single search box. Every finding is a card. Cards open a detail modal. Nothing important lives only in the modal-less page body.
  • Every factual claim carries a clickable source. sources at the bottom is mandatory, and each card that makes a claim gets its own links entry.
  • Files are shown as full absolute paths. Link to GitHub when the file is public.
  • Never fabricate a URL. If you did not open it, do not cite it.

Cards

Group cards so the chips are useful. A good report has 12 to 30 cards across groups such as:

  • Overview — what the thing is, why it matters, the tradeoff space
  • What The Industry Does — mandatory, see below
  • Patterns / Algorithms / Mechanisms — the actual techniques
  • Failure Modes — what breaks in production
  • In This Codebase — real files, real paths, when a repo is involved
  • Recommendations — what to do, ranked

Set groupOrder on the report to the order the sections should render in. Overview first, What The Industry Does second.

Each card needs group, title, summary, tags, and a body of two to five {h, p} blocks. Add bullets, files, and links where they apply. The card teaser is the hook; the modal carries the substance.

Pros, cons and gaps are mandatory

Every card carries:

  • pros — exactly 5 concrete upsides of the thing that card is about
  • cons — exactly 5 concrete downsides
  • gaps — 5 to 10 things it does not solve, does not cover, or leaves open

Rules for these three lists:

  • One line each, no sub-clauses stacked with semicolons. A pro is a claim, not a paragraph.
  • Concrete beats generic. "413k pending pages froze writes" beats "can be slow".
  • A con is not a disguised pro. If you cannot find 5 real downsides, you do not understand the thing well enough yet — go back to the research.
  • Gaps are about the world, not about your report: missing tooling, missing data, unanswered questions, things no vendor does yet, measurements nobody published.
  • Omit gaps only when a card genuinely has none to state. Never omit pros or cons.

What The Industry Does

Every report MUST have a What The Industry Does group. It answers one question: how do the companies with the hardest version of this problem actually solve it?

  • One card per company or lab, with its concrete approach, real numbers, and its own sources.
  • Cover at least 4 of: Google, Amazon/AWS, Meta, Netflix, Stripe, Microsoft/Azure, GitHub, LinkedIn, Cloudflare, Shopify, Uber, GitLab, Salesforce.
  • Cover at least 1 AI lab or AI-platform angle: Anthropic, OpenAI, DeepMind, or the AI platform layer of a cloud vendor, when the topic touches models, agents, search or safety.
  • Include one card naming where the public evidence runs out. Say plainly which companies publish nothing on this and do not invent coverage for them.
  • Vendor docs show the win, incident reports show the cost. Cite both when both exist.

Diagrams

Use handwritten-style inline SVG. templates/handwritten-diagram.svg is the reference: a feTurbulence + feDisplacementMap rough filter, a cursive font stack with real fallbacks, white or #fbeee8 boxes, #1c1b19 strokes.

Diagram rules:

  • Zero crossing arrows. Zero overlapping boxes. If the layout needs a crossing, split it in two diagrams.
  • One flow direction, left to right or top to bottom, never both in one diagram.
  • At most 7 boxes. Number the arrows when order matters.
  • Set an explicit viewBox and leave 20px of margin around the drawing.
  • Skip the diagram entirely rather than shipping a tangled one.

Style

Plain language. No hype, no filler, no "in today's fast-paced world". Short paragraphs. Concrete numbers, limits, and names beat adjectives. State uncertainty when sources disagree.

Final Answer

End with the report path on its own line so the user can click it:

/Users/diegopacheco/git/diegopacheco/html-research/{file}.html

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.