LLMSkillHub
mirroredCopied from solo-agent/solo under MIT. solo-agent has not claimed this page, and is not affiliated with LLM SkillHub.

solo-agent / solo-artifacts

Use when a Solo task or thread should become an interactive, reviewable, self-contained HTML artifact for progress/status, review/decision, or comparison/leaderboard work inside Solo.

0
SKILL.md
3.8 KB

solo-artifacts

Warm Archive fork of work-canvas: produce or publish one self-contained HTML file that renders a Solo task/thread for a human to understand and act on. Prefer the real deliverable over a summary page.

Workflow

  1. Publish Existing Deliverable if the current task/run names one.

    Trust only explicit current-task evidence, such as:

     Deliverable: ./path/to/result.html

    You may also use a file path named in the current task/thread, or a file you personally created for this task now. Do not scan the workspace by newest file, mtime, or vague filename match; old workspace files can be stale. If several files are named, prefer the product/final deliverable over review/report/panel files unless the user explicitly asks for the review page. Do not switch to review-decision just because the task is in_review.

    • Existing .html: publish it directly.

      solo artifact publish --task <task-id> --mode <latest|final> --file <path-to-html>
    • Existing .md, .txt, .json, .csv, or other text/data: create a small self-contained HTML viewer that renders the exact source content, labels the source file near the top, then publish that wrapper. Do not summarize or rewrite the content.
    • Multiple deliverables or a directory: create a small index artifact with explicit links/file names and a short note about what each file is.

    If a real deliverable exists, stop here.

  2. Pick the artifact type and read its reference:

    The task needs to...TypeRead
    Show where long/messy work stands, what broke, what's nextProgress / status reportreferences/progress-report.md
    Review analysis or implementation and decide accept/reject/next stepReview / decision memoreferences/review-decision.md
    Weigh options head-to-head and pick a winnerComparison / leaderboardreferences/comparison.md

    If blended, lead with the primary type and borrow blocks from the other.

  3. Assemble from the template. Copy assets/starter.html, then:

    • Replace [[PASTE base.css]] with the full contents of assets/base.css.
    • Replace [[PASTE interactions.js]] with only the needed modules from assets/interactions.js; pasting the whole file is fine because modules self-guard.
    • Fill the header, What needs your input, and body sections using the component recipes in the chosen reference.
  4. Embed everything: inline CSS/JS, embed media as data: URIs, no CDN or external requests.
  5. Publish back to Solo:

     solo artifact publish --task <task-id> --mode <latest|final> --file <path-to-html>

Non-Negotiables

  • Keep work-canvas structure and interactions intact; use Solo's default Warm Archive visual skin.
  • Real deliverables beat summaries. Never replace an explicit deliverable file with a conversation recap.
  • Surface only real human decisions. If nothing needs the user, say so.
  • Add a legend wherever color, letters, or symbols encode meaning.
  • Never modify Solo/source data from the page. Show paste-ready output with copy buttons.
  • Include a provenance footer with task id, thread/channel scope, agent, model, and generation date when available.

Files

  • assets/starter.html — the skeleton to copy.
  • assets/base.css — class-compatible Solo Warm Archive design system.
  • assets/interactions.js — optional vanilla-JS modules: print, tabs, tables, lightbox, copy, scrollspy, persist.
  • references/*.md — work-canvas section order, component recipes, and guardrails.