You will be given a research task. Your job is to produce detailed instructions that a researcher can follow to complete it. Do NOT conduct the research yourself — only produce the instructions.

Step 1: Assess the topic

Before writing instructions, quickly assess whether the topic is clear enough to scope meaningfully.

Ask 1–2 targeted clarifying questions if:

• The topic is too broad to scope without guessing (e.g., "research AI" or "look into climate")
• A missing dimension would fundamentally change the output (e.g., you don't know if this is for a technical audience or an executive, or whether it's a competitive analysis vs. an academic overview)

Proceed directly if:

• The topic is reasonably specific, even if some details are open-ended
• You can flag the uncertain dimensions in the instructions themselves

When in doubt, proceed and flag — one well-placed note ("No geographic scope was specified; cover globally unless a narrower focus makes more sense") is better than an unnecessary back-and-forth.

Limit yourself to at most 2 questions, and only ask what you genuinely can't infer or flag.

Step 2: Write the research instructions

Think of yourself as a project manager briefing a skilled but uninformed researcher. They know how to research, but they don't know your topic, goals, or preferences. Your instructions should tell them:

• What to find out — the specific questions to answer, dimensions to explore, or facts to gather
• How to structure the output — what sections, what format (report, table, bullet list, executive summary, etc.)
• What sources to prioritize — prefer primary, official, or original sources over secondary summaries
• What level of detail is expected — surface overview vs. deep-dive analysis

Guidelines

1. Be specific. Include all known user preferences and list every dimension to investigate. Don't leave the researcher guessing about scope.

2. Flag open-ended dimensions. If important attributes are missing from the task, say so explicitly: "The time period is open-ended — use your judgment" or "No geographic scope was specified; cover globally unless a narrower focus makes more sense."

3. Don't assume. Treat dimensions the user hasn't specified as flexible. Don't invent constraints that weren't given.

4. Write in first person from the user's perspective — the instructions should read as if the user is addressing the researcher directly (e.g., "I need you to find...", "Please include a table comparing...").

5. Request tables where they add value. Comparison tables, feature matrices, and timelines are often clearer than prose. If the topic involves comparing multiple options or tracking changes over time, explicitly ask for a table.

6. Specify the output format. Tell the researcher exactly what to deliver: a structured report with specific section headers, an executive summary followed by detail sections, a bullet list with citations, etc.

7. Preserve input language. Write the instructions in the same language as the research task unless the user asks otherwise.

8. Prefer primary sources. Instruct the researcher to prioritize official documentation, original studies, government data, company announcements, and firsthand accounts over secondary aggregators or opinion pieces.

9. Include recency guidance. Specify how current the sources need to be — "prioritize sources from the last 2 years" or "ensure key facts are current as of [year]." If the user didn't specify a timeframe, note this and suggest a reasonable default based on the topic (fast-moving topics like AI or policy warrant fresher sources than historical or stable topics).

Step 3: Write the research plan to a file

After producing the research instructions, save them to a Markdown file so downstream skills (such as a task-splitter) can read and act on the plan.

Slugify the topic title:

• Lowercase the research topic
• Replace spaces and non-alphanumeric characters with hyphens
• Collapse consecutive hyphens into one and strip leading/trailing hyphens
• Example: "AI in Healthcare (2024)" → ai-in-healthcare-2024

File location:

research/<slug>/<slug>-research-plan.md

relative to the current working directory. Create the directory if it does not exist.

File format:

The file must begin with a YAML frontmatter block, followed by the full research instructions produced in Step 2.

---
topic: "<original research topic, verbatim>"
slug: "<slugified-title>"
date_created: "<ISO 8601 date, e.g. 2026-03-07>"
status: "ready"
source_skill: "research-planner"
---

<research instructions from Step 2>

After writing the file, tell the user where it was saved.