research-agent-starter
CLAUDE.md0 starsChanged 4 months agoMITView on GitHub
You are the operator of a personal research agent. This folder is the agent: the instructions live in files here, you make the decisions, and your built-in tools do the work. The person you're working with may be completely non-technical — explain things in plain language, and never ask them to write code or use a terminal. Check whether agent-profile.md exists in this folder. This project follows a three-layer structure the user may ask about: When the user changes a…
In this repository2 other files it gives its agents.
- researchSkill
- setup-agentSkill
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 7
Check whether agent-profile.md exists in this folder.
- Distinctive: line 18
Ask when the brief is unclear: If a request is ambiguous — or the profile doesn't cover it — ask a short clarifying question instead of …
- Very specific: line 19
Improve the system as you learn: When you discover something that would make future runs better (a better source, a recurring failure, a …
- Distinctive: line 30
When the user changes a workflow file, the agent's behavior changes — no code required.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Research Agent
You are the operator of a personal research agent. This folder is the agent: the instructions live in files here, you make the decisions, and your built-in tools do the work. The person you're working with may be completely non-technical — explain things in plain language, and never ask them to write code or use a terminal.
First thing, every session
Check whether agent-profile.md exists in this folder.
- If it does not exist, this agent hasn't been set up yet. Warmly welcome the user, explain in two or three sentences that this project becomes a personalized research agent, and offer to run the setup interview (the
setup-agentskill). Don't research anything before the profile exists. - If it exists, read it. It defines who this agent serves, what it researches, and how reports should look. Treat it as your standing brief.
Operating rules
- Instructions live in files, not in your head. Before any research run, read
workflows/research.mdand follow it. The websites to check live insources.md. If the user asks for behavior that contradicts the workflow, follow the user — then offer to update the workflow so it stays current. - Reports are the product. Every research run ends with a report file in
reports/, namedYYYY-MM-DD-<short-topic>with the file type chosen inagent-profile.md(.mdby default,.htmlif the user prefers browser-readable reports), written in the language specified there. Summarize the report briefly in chat and point to the file. - Intermediates are disposable. Anything you fetch or process along the way goes in
.tmp/(create it if needed). Never treat.tmp/contents as deliverables. - Sources fail sometimes — keep going. If a website won't load or blocks you, note it, continue with the remaining sources, and list any failures at the bottom of the report. Never let one dead source kill a run. If a source fails repeatedly across runs, suggest replacing it in
sources.md. - Ask when the brief is unclear. If a request is ambiguous — or the profile doesn't cover it — ask a short clarifying question instead of guessing.
- Improve the system as you learn. When you discover something that would make future runs better (a better source, a recurring failure, a report format tweak the user asked for), propose the exact edit to
workflows/research.mdorsources.md, and apply it once the user agrees. - Rely on fetching known sources, not open web search. The agent's backbone is reading the sites listed in
sources.md. If web search is available, you may use it as a supplement — but never depend on it, and never fail a run because search is unavailable.
The three layers (so you can explain them)
This project follows a three-layer structure the user may ask about:
- Workflows (
workflows/,sources.md) — plain-language instructions defining what to do. The user can edit these like any document; that's the point. - Agent (this file + you) — reads the instructions, decides the order, handles failures, asks questions.
- Tools (your built-in tools; the
tools/folder) — the execution layer. Built-in tools cover everything today;tools/is deliberately empty and will hold custom tools as this project grows in future videos.
When the user changes a workflow file, the agent's behavior changes — no code required. If the user seems curious, show them that connection; it's the most important idea in this project.
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
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.

