agentleFS
buildwithdorghambuildwithdorgham/research-agent-starter/.claude/skills/setup-agent/SKILL.md

setup-agent

Skill0 starsChanged 4 months agoMITView on GitHub

Use when the user wants to set up, customize, retarget, or re-configure this research agent — including first-time setup (no agent-profile.md exists yet), pointing the agent at a new research goal, or requests like "set up my agent", "customize this for me", "I want to research something else", or "run the setup interview".

3 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions3 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 3

    1. show-karma/skills★ 28same name
    2. coval-ai/coval-external-skills★ 3same name
    3. logan-bishop-sgs/sgs-agent-skills★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If agent-profile.md already exists, tell the user this will retarget their agent: the profile, workflow, and source list get rewritten …

    line 12
  2. Distinctive:

    Ask the questions one at a time.

    line 13
  3. Distinctive:

    Write the step-by-step procedure for a research run, tailored to their answers.

    line 64

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Setup Interview — build the agent around the user

You are about to turn this empty framework into the user's personal research agent. You'll interview them, then write three files that define the agent. The user may be non-technical: keep every question short, offer examples, and never use jargon.

Before you start

  • If agent-profile.md already exists, tell the user this will retarget their agent: the profile, workflow, and source list get rewritten around the new goal. Confirm before proceeding. Offer to save a copy of the old files into .tmp/previous-setup/ first.
  • Ask the questions one at a time. Wait for each answer. Keep the pace of a friendly conversation, not a form.

The interview

  1. What do you want this agent to research? Offer examples: competitors in your market · a business idea you're validating · a market or niche · the latest AI news · anything else they name.
  2. What decision will this research help you make? This is the most important question — it turns "interesting information" into "useful information". Examples: "decide my product's pricing", "decide whether to enter this market", "stay current so I can create content about AI".
  3. Who will read the reports? Just them? A client? A team?
  4. Which sources do you trust for this topic? First ask if they have websites they already follow. Then propose a starter list of 4–6 well-known sources that fit their topic, and let them add, remove, or replace freely. The final list is theirs, not yours.
  5. How often will they run this? One-off deep dive, or recurring (daily / weekly / monthly)? This shapes the workflow: recurring runs should emphasize what's new since last time; one-off runs should be comprehensive.
  6. What should a report look like? A short executive brief, a bullet-point digest, or a deep report? Roughly how long? Anything that must always appear (e.g. sources cited, a "so what" section, action recommendations)?
  7. What file type should reports be saved as? Two good options — don't assume, ask:
    • Markdown (.md) — the default; clean text that reads beautifully inside Claude and on GitHub.
    • HTML (.html) — opens styled in any web browser and prints/saves to PDF easily; the better fit if they'll share reports or read them outside Claude. (For Arabic reports, mention that HTML displays right-to-left properly.)
  8. What language should reports be written in? The project files are in English, but reports belong to the user — Arabic, English, or anything else they prefer.
  9. Anything the agent should always ignore or never include? Examples: paywalled content, rumors and speculation, topics out of scope, a competitor they don't care about.

After the interview

Summarize the answers in a few lines and confirm before writing anything. Then write these three files:

1. agent-profile.md

# Agent Profile

- **Research focus:** <their topic, specific>
- **Decision it feeds:** <their answer>
- **Audience:** <who reads the reports>
- **Cadence:** <one-off / recurring + frequency>
- **Report language:** <language>
- **Report format:** <format + target length + must-have sections>
- **Report file type:** <Markdown (.md) or HTML (.html)>
- **Always ignore:** <their exclusions, or "nothing specified">
- **Set up:** <today's date> — re-run /setup-agent anytime to retarget this agent.

2. workflows/research.md

Write the step-by-step procedure for a research run, tailored to their answers. Use this skeleton, filled in with their specifics — plain language, numbered steps:

# Research Workflow

**Objective:** <one sentence: what a run produces and for whom>

## Steps

1. Read `agent-profile.md` and `sources.md`.
2. If the user gave a specific topic for this run, focus on it; otherwise cover the standing focus above.
3. Fetch each source in `sources.md`. Save anything worth keeping to `.tmp/`.
   - If a source fails, note it and continue. List failures at the end of the report.
4. <For recurring agents: compare against the most recent report in `reports/` and prioritize what is NEW since then.>
5. Synthesize — don't just list. Group findings by theme, and connect them to the decision this research feeds: <their decision>.
6. Write the report to `reports/YYYY-MM-DD-<short-topic>.<md or html, per the profile's file type>` in <language>, following the format below.
7. Tell the user the report is ready, with a 3-line summary in chat.

## Report format

<their format: sections, length, must-haves — spelled out concretely>

## Edge cases

- Fewer than half the sources loaded → tell the user the run is thin and offer to retry later rather than presenting weak findings as complete.
- A finding is rumor or speculation → either exclude it or clearly label it, per the profile's exclusions.

3. sources.md

The confirmed source list, in this format:

# Sources

<!-- One source per line. The agent checks these on every run. Edit freely — this file is yours. -->

- <Source name> — <URL> — <one-line note on why it's trusted / what it's good for>

Verify the sources before finishing

Some websites block automated reading — better to find out now than during the user's first report. After writing sources.md, quickly test-fetch each source. If one fails, tell the user, suggest a working alternative for the same kind of coverage, and update the list with whatever they choose. Ship only sources that actually load.

Finish

Tell the user their agent is ready, name the three files you wrote (so they see where their answers went), and invite them to run their first report with /research. One sentence, not a lecture.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.