
write-article
by feixeyes
Using large models to do real things.
SKILL.md
name: write-article description: End-to-end workflow for writing an article in the Real World AI Log project, focusing on practice-first, workflow-first collaboration. version: 0.1
write-article
This skill manages the full article lifecycle for the project AI 实践记录 / Real World AI Log.
The goal is not to maximize writing speed, but to enforce a clear, reviewable, and repeatable workflow.
Agents using this skill must behave as collaborators, not autonomous authors.
When to Use
Use this skill when:
- Starting a new article from an idea or rough topic
- Turning an experience or practice into a structured article
- Managing multi-step writing with explicit human checkpoints
Do NOT use this skill for:
- Minor edits or copy changes
- Style-only polishing
- One-off brainstorming
Core Philosophy
- Practice > theory
- Process > polished conclusions
- Human judgment > agent autonomy
- Small steps > big jumps
The agent must never skip planning or confirmation steps.
Required Workflow
Phase 1: Context Intake (Mandatory)
Before proposing any outline or text, the agent MUST:
-
Read
README.md -
Read
AGENTS.md -
Scan the latest 2–3 files in
content/published/ -
Identify:
- Target audience (explicit or assumed)
- Relevant column/series (if any)
- Tone constraints and boundaries
The agent must summarize its understanding in 5–7 bullet points.
STOP and wait for confirmation if understanding is unclear.
Phase 2: Planning & Outline Proposal (Mandatory)
The agent MUST propose 2–3 outline options, each including:
- Central question or problem statement
- Intended reader takeaway
- High-level section list
- Trade-offs (why choose / not choose this outline)
The agent MUST:
- Explicitly ask the human to choose one option
- NOT proceed until a choice is confirmed
Phase 3: Task Breakdown (Mandatory)
After an outline is confirmed, the agent MUST:
-
Break the work into small tasks, typically:
- Create outline file →
content/outlines/<slug>.md - Create draft file →
content/drafts/<slug>.md
- Create outline file →
-
Describe what each task will produce
-
Ask for confirmation before executing tasks
CRITICAL TASK FLOW:
- Task 1 (Create outline file): Execute, then WAIT for user confirmation or manual edits
- Task 2 (Create draft file): Execute ONLY AFTER user explicitly confirms the outline is finalized
No task execution without approval.
Phase 4: Execution (Controlled)
For each approved task:
- Execute one task at a time
- Produce a concrete artifact in the correct folder
- Summarize what was done
- Ask whether to continue to the next task
Do NOT chain tasks without explicit approval.
Phase 5: Self-Review & Checks
After draft completion, the agent MUST self-check for:
- Logical flow and clarity
- Overconfident or authoritative tone
- Missing concrete examples
- Redundancy or unnecessary abstraction
- Alignment with project principles
The agent should output:
- A short review summary
- A list of suggested human edits (not applied automatically)
Writing Constraints
The agent MUST follow these constraints:
- No hype or exaggerated AI claims
- Prefer first-person reflective tone when appropriate
- Avoid tutorial-style commands
- Use short to medium paragraphs
- Do not pretend to have final answers
Guardrails (Hard Rules)
- ❌ Do not start writing without an approved outline
- ❌ Do not publish or move files to
published/ - ❌ Do not invent facts or experiences
- ❌ Do not overwrite existing files without permission
If uncertain, STOP and ask.
Success Criteria
This skill is successful when:
- The human can easily review, modify, or stop the process
- Each step produces a clear, inspectable artifact
- The workflow reduces cognitive load for future writing
- The result feels honest, grounded, and practice-driven
Consistency and clarity are more important than speed.
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon