dk-wiki-article-writing

Files (11)

• SKILL.md
• 38e8556bf37cbabaf47cf68ad39f45fbdcb1bb0d72a7457f523c33b27df404a3 · apicache/
• 8a5d7886193b84bdb43d4b401b5168a1c89d1d7f66722482e38477f7a4484f9d · apicache/
• 992472921723f31c65c963f1aeb1c7af0574c3ca8538f694076afd4e6e5ca62d · apicache/
• 9f4738a9683ad8317eccf68a9f0ac99733ad245f8591f27775196ff85281262f · apicache/
• f7a412aba759ec2c6dad43e89bf027338c17aecebb83be97016bd3317607ff20 · apicache/
• html-template.md · references/
• search.py · scripts/
• search.sh · scripts/
• wiki.py · scripts/
• throttle.ctrl

SKILL.md (first lines)

# Wiki Article Writing

A research-first workflow for writing comprehensive, balanced, Wikipedia-style articles on technical topics. The core principle: **research breadth before writing depth** — never start writing until you've heard from voices you didn't expect to find.

## Overview

This skill enforces a 6-phase pipeline that prevents the most common failure mode in AI-assisted article writing: **advocacy bias from narrow sourcing**. When you only search for the names you already know, you get a highlight reel, not an encyclopedia.

Research is parallelised via subagents — multiple research threads run concurrently, each writing findings to files. Synthesis happens only after all threads complete.

## When to Use

- Writing a Wikipedia-style article on a technical topic
- Creating a comprehensive research document that needs balanced coverage
- Producing an HTML page displaying a well-sourced article
- Any task requiring multi-source research synthesis before writing

## When NOT to Use

- Quick summaries or opinion pieces (no balance needed)
- Internal documentation (no citation requirements)
- Topics with no existing discourse (nothing to balance against)

---

## Phase 0: Research Plan

**Goal:** Define what to research before searching. This prevents reactive, unfocused searching.

1. **Create the research folder:**

```bash
mkdir -p research_{topic_slug}
```

2. **Write `research_{topic_slug}/research_plan.md`** containing:

```markdown
# Research Plan: {Topic}

## Main Research Question
{What is this article about?}

## Subtopics (3-5, non-overlapping)

### Subtopic 1: {name}
- **Search angles:** {2-3 specific queries}
- **Expected sources:** {what kind of blogs, docs, discussions}
- **Budget:** 3-5 searches max

### Subtopic 2: {name}
- **Search angles:** {2-3 specific queries}
- **Expected sources:** {what kind of blogs, docs, discussions}
- **Budget:** 3-5 searches max

### Subtopic 3: {name}
...

## Synthesis Approach
How will findings combine into a balanced article?
```

**Subtopic sizing guide:**

| Topic complexity | Subtopics | Parallel subagents |
|-----------------|-----------|-------------------|
| Simple fact-finding | 2 | 2 |
| Comparative analysis | 3 | 3 |
| Complex multi-perspective | 4-5 | 3 (batch remaining) |

**Save to:** `research_{topic_slug}/research_plan.md`

---

## Phase 1: Wikipedia Baseline

**Goal:** Understand what already exists and mine references.

Fetch the Wikipedia article:

```bash
uv run {baseDir}/scripts/wiki.py "YOUR_TOPIC_HERE" --full --refs --toc
```

`uv run` auto-installs pywikibot on first run. Outputs article text, reference URLs, and TOC links.

**Note gaps:** What's missing? What's underdeveloped? What's the tone?

**Save to:** `research_{topic_slug}/wikipedia-baseline.md`

---

## Phase 2: Parallel Research (Subagents)

**Goal:** Find voices across the full spectrum — proponents, critics, practitioners, independent observers — efficiently via parallel delegation.

### Delegation Strategy

Launch all research subagents in a **single `subagent({ tasks: [...] })` call** using PARALLEL mode. Do not spawn them one at a time — the system only allows one subagent invocation per turn, so parallel mode is the only way to run concurrent research.

**Subagent task template:**

```
Research "{SUBTOPIC_NAME}" for a Wikipedia-style article about {TOPIC}.

Search angles:
- {query_1}
- {query_2}
- {query_3}

Use the search tool:
  {baseDir}/scripts/search.sh "{query}" 10

For the 2-3 most important results, fetch full content (captures nuance that snippets miss):
  {baseDir}/scripts/search.py "{url}" -n 1 --content

CRITICAL: You MUST write your findings to research_{topic_slug}/findings_{subtopic_slug}.md using the write tool BEFORE finishing your session. Do not end without writing this file.

Write your findings with:

From dushyantkhosla/agent-skills, MIT.

More from dushyantkhosla/agent-skills