agentleFS
ai-ecoverseai-ecoverse/skills/skills/speck/SKILL.md

speck

Skill1 starsChanged 30 days agoApache-2.0View on GitHub

Use this when the user wants to annotate, inspect, or leave element-level feedback on a local HTML page served in a browser tab. Injects a hover-highlight + click-to-annotate interaction into the target page. Annotations are sent as licks to a scoop that applies the user's instructions as AI-driven edits to the HTML. Use when the user says "annotate this page", "let me mark up elements", "add speck", "inspect elements with notes", or "inject speck".

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before speck can work end-to-end, the cone must ensure:

    line 24
  2. Distinctive:

    Read the target file

    line 31
  3. Very specific:

    When the cone receives [@speck-worker-scoop completed]:

    line 108
  4. Distinctive:

    Read the notification preview to see what was applied.

    line 109
  5. Distinctive:

    Don't inject into remote/third-party pages (CSP will block it).

    line 114
  6. Very specific:

    Don't call speck inject twice on the same tab without speck remove first — you'll get duplicate listeners.

    line 116
  7. Very specific:

    Don't create multiple webhooks — speck inject looks for an existing speck-lick webhook before creating one.

    line 119
  8. Very specific:

    Don't use shift-reload on the preview page — it bypasses the service worker and gives ERR_FILE_NOT_FOUND.

    line 121
  9. Distinctive:

    Don't create a sprinkle for speck progress — it was tried and removed because the rail icon can't show dynamic state.

    line 123

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

speck

Inject an element-level annotation layer into any local HTML page currently open in a browser tab. Once active, hovering highlights elements with a coloured border; clicking locks the selection and opens a text input where the user can type an instruction for that element. On submit, the instruction is sent as a webhook lick to a speck-worker scoop that applies the edit to the underlying HTML file, reloads the page, and re-injects speck for continuous editing.

Setup (cone responsibilities)

Before speck can work end-to-end, the cone must ensure:

  1. A speck-worker scoop exists with /tmp/ write access:
    scoop_scoop("speck-worker", writablePaths=["/scoops/speck-worker-scoop/", "/shared/", "/tmp/"])
  2. The scoop has standing instructions (via feed_scoop) to:
    • Read the target file
    • Locate the element by selector/snippet
    • Apply the user's instruction as an edit
    • Reload the tab via playwright-cli goto <preview-url> --tab <tabId>
    • Re-inject speck via speck inject <tabId> --file <filePath>
  3. A single webhook named speck-lick routes to the scoop. The speck inject command creates this automatically if it doesn't exist.

Commands

speck inject <tabId> [--file <path>]   # Inject + wire webhook for live edits
speck collect <tabId>                  # Retrieve annotations as JSON
speck remove <tabId>                   # Remove overlay + cleanup state

The --file flag tells the speck-worker scoop which HTML file to edit. Without it, annotations are collected locally but not applied.

How it works

  1. Hover: elements receive an orange outline (3px solid #FF6B35) with a soft glow on mouseover.
  2. Click: locks the highlight and shows a fixed-position dark input tooltip below the element.
  3. Enter: submits the instruction — POSTs to the speck-lick webhook which routes to the speck-worker scoop. The tooltip flashes green to confirm.
  4. Escape: dismisses without submitting. Click elsewhere also dismisses.
  5. Scoop execution: the scoop edits the file, reloads the page, and re-injects speck so the user can keep annotating without interruption.

Webhook payload shape

{
  "action": "element-instruction",
  "data": {
    "instruction": "make this text larger",
    "element": {
      "tag": "h2",
      "class": "section-title",
      "text": "Latest News",
      "selector": "section:nth-of-type(3) > h2",
      "outerSnippet": "<h2 class=\"section-title\">Latest News</h2>"
    },
    "file": "/shared/stardust/prototype/index.html",
    "tabId": "1218556234",
    "timestamp": 1717300000000
  }
}

Progress feedback

The cone sends a chat message when the speck-worker scoop completes each edit, summarizing what was applied. This gives the user immediate feedback in the conversation without needing a separate UI panel.

Key design decisions

  • Single webhook: speck inject reuses a single webhook named speck-lick rather than creating one per tab/injection. This prevents duplicate webhook accumulation.
  • Re-injection after edit: the scoop MUST re-inject speck after reloading the page, otherwise the overlay is lost.
  • No sprinkle UI: progress is shown via chat messages from the cone. A sprinkle panel was tried but the rail icon cannot show dynamic status, making it less useful than inline chat feedback.
  • Preview pages only: the injection uses playwright-cli eval on the extension's preview service worker pages. Remote/third-party pages will block it via CSP.
  • /tmp/ write access: the scoop needs /tmp/ to run speck inject (which writes a temp file for playwright-cli eval). Include it in writablePaths when creating the scoop.

Cone workflow on scoop completion

When the cone receives [@speck-worker-scoop completed]:

  1. Read the notification preview to see what was applied.
  2. Send a short chat message to the user: "Speck applied: <summary>"

Don't

  • Don't inject into remote/third-party pages (CSP will block it). Use only on locally served or preview pages.
  • Don't call speck inject twice on the same tab without speck remove first — you'll get duplicate listeners. (The script guards against this with an already injected check, but avoid it anyway.)
  • Don't create multiple webhooks — speck inject looks for an existing speck-lick webhook before creating one.
  • Don't use shift-reload on the preview page — it bypasses the service worker and gives ERR_FILE_NOT_FOUND. Use regular reload or playwright-cli goto.
  • Don't create a sprinkle for speck progress — it was tried and removed because the rail icon can't show dynamic state.

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.