AGENT SKILL

editorial-document

Write long-form editorial documents as clean, semantic HTML saved as an artifact (kind=document) — reports, briefs, explainers, proposals, write-ups. Covers the house style (structure, voice, typography) AND the security contract (the HTML is sanitized fail-closed, so author within the allowlist). Load when the user wants a polished written document, report, brief, or explainer that should read like a publication, not a chat message.

License 许可
Apache-2.0
最近更新
2026-11-05
Stars
★ 212
Forks
33

01 前言 Front Matter

name
editorial-document
triggers
document, write a report, write up, brief, explainer, proposal, white paper, editorial, long-form, publish a doc, formatted document, article, memo, one-pager
description
Write long-form editorial documents as clean, semantic HTML saved as an artifact (kind=document) — reports, briefs, explainers, proposals, write-ups. Covers the house style (structure, voice, typography) AND the security contract (the HTML is sanitized fail-closed, so author within the allowlist). Load when the user wants a polished written document, report, brief, or explainer that should read like a publication, not a chat message.

03 SKILL.md 正文

Editorial Documents

Gideon renders long-form written documents from semantic HTML, saved as an artifact with kind=document. Unlike a widget (interactive HTML in a sandboxed iframe) or markdown, a document is prose-first editorial HTML rendered in the page with a tuned reading type scale — it should read like a published piece.

Save with artifact_save(kind="document", name="…", content=<HTML>). The body is HTML (semantic tags, no <html>/<body> wrapper needed — just the content). The viewer renders it with the editorial stylesheet, lets the user edit the HTML with a live split preview, comment on any passage, and export it (standalone HTML / copy).

When to use this vs other types

  • Document — a written piece meant to be read: reports, briefs, explainers, proposals, post-mortems, design write-ups. Prose is the substance.
  • Markdown — quick notes / READMEs / chat-adjacent text where Markdown's shorthand is enough and you don't need editorial typography.
  • Widget (visual-output skill) — interactive HTML, dashboards, custom charts.
  • Infographic (infographic-syntax skill) — structured visual info-design.

The security contract (author within the allowlist)

The document body is sanitized fail-closed before rendering: anything not on the allowlist is dropped. This is a security control because the content may echo untrusted material (e.g. crawled pages). Author accordingly:

  • Allowed: semantic prose + structure — headings (h1–h6), p, ul/ol/ li, blockquote, pre/code, table/thead/tbody/tr/th/td, figure/figcaption, img, a, strong/em, hr, section/article, inline svg for simple diagrams.
  • Dropped (don't bother emitting): <script>, <style>, <iframe>, <object>, <form>, inline event handlers (onclick=…), inline style= attributes, and javascript:/data:text/html URLs. Links must be http(s), mailto:, tel:, relative, or #anchors; data:image/* is allowed for inline images.
  • Don't fight the sanitizer with inline styles — the editorial stylesheet handles typography. Write clean structure and let it style.

House style

Structure. Open with an <h1> title and a one-paragraph lede that states the point. Use <h2> for major sections, <h3> for sub-sections. Prefer short sections with descriptive headings over one long wall. Close with a clear takeaway or next-steps section when the piece calls for it.

Voice. Direct and concrete. Lead with the conclusion, then support it (BLUF). Active voice. Cut filler ("it should be noted that", "in order to"). One idea per paragraph; 2–4 sentences each.

Evidence. Use <table> for comparisons/data, <ul> for parallel points, a <blockquote> for a pulled quote or key caveat. Link sources with real <a href>. Use <figure> + <figcaption> for images/diagrams.

Typography. Don't hand-set fonts/sizes/colors — the reader stylesheet owns the type scale (heading hierarchy, a measure of roughly 75 characters, readable line-height) and is theme-aware. Your job is correct semantics; the styling follows.

Example

<h1>Q3 Reliability Review</h1>
<p>Availability held at 99.95% despite a 40% traffic increase — but two incidents
share a root cause we haven't fixed. This brief lays out what happened and the one
change that closes both.</p>

<h2>What happened</h2>
<p>…</p>
<table>
  <thead><tr><th>Incident</th><th>Duration</th><th>Root cause</th></tr></thead>
  <tbody>
    <tr><td>Aug 12 latency</td><td>34 min</td><td>Connection-pool exhaustion</td></tr>
    <tr><td>Sep 3 errors</td><td>18 min</td><td>Connection-pool exhaustion</td></tr>
  </tbody>
</table>

<h2>The fix</h2>
<blockquote>One change — bounded pools with backpressure — removes the shared cause.</blockquote>
<p>…</p>

Workflow

  1. Confirm a document is right (a written piece to be read, not interactive).
  2. Draft semantic HTML within the allowlist; structure with headings + the lede.
  3. artifact_save(kind="document", name="…", content=<HTML>) (or artifact_update a prior one — artifact_list first, per the artifacts skill).
  4. The user reads it, edits the HTML with live preview, comments on passages (those route back to you), and can export a standalone HTML file.

Pitfalls

  • Don't emit <script>/<style>/inline style=/onclick= — they're stripped.
  • Don't wrap the body in <widget> — kind=document is its own type.
  • Don't hand-style — semantics + the reader stylesheet, not inline CSS.
  • Keep links to safe schemes (http(s)/mailto/tel/relative/#).

04 同仓库更多 Skills

Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

web-verify

Verify a built web page or app actually renders and works — drive it in a real browser, check the console and network, confirm the thing you changed is the thing being served, and capture evidence. Use before claiming a frontend change works.

★ 212
Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

visual-output

Emit rich visual output in chat via <widget> tags — styled HTML cards, charts, tables, interactive tools, and drawn SVG/canvas diagrams, illustrations, and animations — theme-aware, centered, and responsive instead of clashing or cramped.

★ 212
Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

task-and-project

Track work as Tasks (the /api/tasks CRUD + native task provider), and decompose larger projects into independent, verifiable units — fanning parallel investigation out to subagents while keeping one coherent plan; hand open-ended goal-driven work to a goal loop.

★ 212
Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

research-campaign

Run a multi-cycle research campaign — grill the question, decompose it into answerable sub-questions, investigate each against sources, then synthesize with confidence and gaps stated. Use for open questions that need real investigation rather than one lookup.

★ 212
Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

memory-discipline

When and how to persist durable memory — save genuine user preferences, facts, and corrections; NEVER persist transient environment failures (a failed command, a flaky network call, a one-off tool error) as lessons; recall before re-asking.

★ 212
Sidiora-Labs
Sidiora-Labs/centra-gideon-agent

loop-worker

Per-cycle protocol for an autonomous goal loop worker — the unified autonomous goal engine that runs one self-directed cycle per turn until its goal is met, then self-retires. Drives one cycle against a loop's file interface (status.json gate, brief.md goal, guidance.txt nudges, findings/cycle_NNN.json, FINDINGS.md, the type-appropriate deliverable, STOP sentinel). The worker produces work and reports evidence; a deterministic check or a separate judge — never the worker — decides done-ness.

★ 212