Skip to content

SKILL.md

Latest commit

 

History

History
179 lines (141 loc) · 7.32 KB

SKILL.md

File metadata and controls

179 lines (141 loc) · 7.32 KB
Raw
name linkyee-plugin-builder
description Use when the user wants to add dynamic data (GitHub stars, latest blog posts, weather, follower counts, repo activity, anything fetched from a URL) to their linkyee site by writing a build-time plugin. Triggers on phrases like "add a plugin", "show my latest Medium post on the page", "fetch X and display it", "make linkyee pull data from Y", "inject a value into the page". Generates `plugins/<PluginName>.rb`, wires it up in `config.yml`, references it in the right Liquid spot, and runs the build to verify.
allowed-tools Read, Write, Edit, Bash, Glob, Grep

linkyee Plugin Builder

You help users add build-time plugins to linkyee — a Ruby/Liquid static-site generator. A plugin is a small Ruby class that fetches some data at build time and exposes it to the templates as vars.<PluginName>.

The canonical reference for the plugin contract lives at plugins/README.md. Read it before you write code; it contains examples, helpers, and conventions that this skill expects you to follow. Do not duplicate that information here — this skill is the workflow, the wiki is the spec.

Mental model in one paragraph

config.yml lists enabled plugins. At build time scaffold.rb instantiates <PluginName>.new(yaml_values), calls execute(), and stores the return value at settings["vars"][<PluginName>]. Liquid then renders {{ vars.<PluginName> }} (and friends) inside config.yml strings and the theme's index.html. If a plugin raises, scaffold logs the error and sets the value to nil — the build still completes.

Workflow

Follow these steps in order. Don't skip steps; the verification step in particular catches most mistakes.

1. Understand the request

Before writing anything, confirm:

  • What data does the user want? (a star count, a list of posts, a weather temp, a follower count, the latest commit, …)
  • From where? (specific URL, RSS feed, JSON API, scraped HTML)
  • Where on the page should it appear? (footer, a link's text, a new link, a new section in index.html)

If anything is ambiguous, ask one focused question. Don't ask three clarifying questions at once.

2. Check whether a built-in plugin already covers it

Read plugins/README.md § "Built-in plugins". If GithubRepoStarsCountPlugin, GithubLastCommitPlugin, GithubProfilePlugin, or RSSFeedPlugin already does the job, don't write a new one — just enable and reference it. Tell the user.

3. Read the existing config

Run Read on config.yml to see how the user already structures their links and what's already in vars. This affects naming and where to inject the value.

4. Write the plugin

Create plugins/<PluginName>.rb:

  • Name the class to match the file. Use a descriptive *Plugin suffix (e.g. WeatherPlugin, MediumLatestPostsPlugin).
  • require_relative 'Plugin' at the top.
  • Inherit from Plugin.
  • Use the helpers (http_get, http_get_json, args, params, cache, log) — do not call Net::HTTP or URI directly. They're already wrapped with redirect following, timeouts, and error handling.
  • Be defensive. Return a sensible default (0, "", {}, []) on failure. Plugins must not raise.
  • Use String keys in returned hashes (Liquid can't look up symbol keys).
  • Match the data shape on success and failure (don't return a Hash on success and nil on failure — Liquid breaks).

Look at the existing built-in plugins as templates — they're short and demonstrate the patterns.

5. Wire it up in config.yml

Add to plugins: using the YAML style that matches the plugin's arguments (list-style for args, hash-style for params). Then reference {{ vars.<PluginName>… }} in the appropriate links/socials/title/tagline/footer field, or edit the theme's index.html if the user wants a loop / conditional.

6. Verify the build

Run:

bundle exec ruby ./scaffold.rb

Then check:

  • Build exits 0
  • No [<PluginName>] or [scaffold] Plugin '<PluginName>' failed lines in stderr
  • The injected value appears in _output/index.html (grep for it)

If the value is missing, read the stderr output — the new scaffold.rb swallows plugin failures but logs them clearly. Do not guess; fix the underlying error.

7. Secrets — non-negotiable

If the plugin needs an API key, personal access token, OAuth token, or any other credential:

  • Never put the secret value in config.yml. That file is committed to git and rendered into the public site at build time. Anything in it is public, forever, including in old commits.
  • Never hardcode the secret in the plugin's .rb file. Same problem.
  • Always read it from ENV: 
    token = ENV["MEDIUM_TOKEN"]
return [] if token.nil? || token.empty?
http_get_json(url, headers: { "Authorization" => "Bearer #{token}" })
  • Always document the secret name in the plugin's header comment so the user knows what to set.
  • Always tell the user how to wire it through, in three steps:
    1. Repo secret: GitHub → repo → Settings → Secrets and variables → Actions → New repository secret. Name it (e.g. MEDIUM_TOKEN), paste value.
    2. Workflow env: edit .github/workflows/build.yml, add an env: block to the Deploy step: 
      - name: Deploy
  env:
    MEDIUM_TOKEN: ${{ secrets.MEDIUM_TOKEN }}
  run: bash deploy.sh
    3. Local dev: same name, exported in shell: export MEDIUM_TOKEN=xxx && bundle exec ruby ./scaffold.rb

If the user asks you to "just paste the token in config.yml for now", refuse and explain why. The secret will end up in git history and on the deployed gh-pages branch — no easy way to revoke once leaked.

8. Report back

Tell the user:

  • What plugin was created
  • What got added to config.yml
  • Where in the rendered page the value now appears
  • How to extend it (e.g. "add more entries to the list under MyPlugin:")

What NOT to do

  • Don't add new gems to the Gemfile unless absolutely required. nokogiri (already in Gemfile) plus stdlib net/http and json cover the vast majority of cases.
  • Don't scrape sites that have an obvious public JSON/Atom endpoint. RSS / Atom (/feed, /commits.atom) and JSON APIs are far more stable than HTML scraping.
  • Don't hardcode credentials. If a token is required, read it from ENV and document the secret name in the plugin's header comment.
  • Don't add per-request side effects (writing files, sending notifications). execute should be a pure data fetch.
  • Don't invoke things at build time that can take more than ~30s in total — the GitHub Actions build is on a free runner and we run it daily. Cap the work.
  • Don't touch theme files unless the user asked for layout changes. Most plugin requests are satisfied by editing config.yml only.
  • Don't create a new plugin when an existing one (RSSFeedPlugin especially) already handles the case.

Anti-patterns to refuse

  • "Make it run in the visitor's browser instead." — linkyee is a static site; plugins are build-time only. If they want runtime data, they need separate JS in the theme — out of scope for this skill.
  • "Have the plugin email me when X." — no side effects in plugins. Suggest a separate GitHub Actions workflow.