My first agent

Intro

I've been learning to use Claude Code for data science projects, and my first real test was building a simple agent from scratch — starting with a CLAUDE.md file that defines how the agent should behave. Along the way I hit a classic gotcha (files quietly saving to a temp folder instead of my project directory), fixed it, and then put the whole setup through a real research task. Here's what I learned.

Building a Simple Agent with Claude Code (VS Code)

The prompt is saved in a CLAUDE.md file, with the following content:

# try to build an agent
## Project Context
This is an AI agent playground.

## About me
I'm a DS, I'm learning to use Claude Code for DS projects.

## Rules
- Always ask clarifying questions before starting a complex task
- Show your plan and steps before execution
- Keep reports and summaries concise — bullet points over paragraphs
- Save all reports to the output folder
- Cite sources when doing research
## Project Structure
- `resources/` -- reference docs and templates
- `workflows/` -- workflow instruction files (plain English recipes the agent follows)

## Output Rules
- All deliverables MUST be saved under the absolute path: `C:\claudecoderepo\my_first_agent\output`
- Never write files to system temp directories or any path outside this project folder
- Before saving, confirm the current working directory matches the project root

Testing the Agent: A Real Research Task

Next, I opened Claude Code's planning mode with Shift + Tab and gave it a real task:

  • Web search data scientist position requirements in 2026 and compare them to 2025 — are there signs of increasing demand for AI knowledge?

  • Is the trend in job openings increasing or decreasing?

  • Also, web search efficient strategies for landing a DS job in 2026.

  • Save the results — for the data scientist position research only — to the specified output folder.

Claude Code executed the task step by step, pausing to ask for confirmation at each stage, exactly as defined in the CLAUDE.md rules. It then saved the final report, Data Scientist Job Market Research (2026).md, to the target output folder — which worked flawlessly.

I've included the full report in another post.

Takeaways

A few things stood out from this first pass:

  • Relative paths are a trap. "Save to output/" sounds unambiguous to a human, but Claude Code resolves it against whatever directory it thinks it's in. Spelling out an absolute path in CLAUDE.md fixed the issue for good.

  • A CLAUDE.md file is a contract, not a suggestion. Once I made the rules explicit and specific (confirm working directory, ask before complex tasks, cite sources), the agent actually followed them — down to pausing for confirmation at each step.

  • Planning mode (Shift + Tab) is worth using for anything multi-step. Watching the agent lay out its plan before executing made it much easier to trust the output, especially for a research task with several sub-parts.

This is a real shift from how I worked before. For my CFPB project (see previous post), my workflow with Claude was entirely conversation-based: I'd chat with Claude, copy the code it generated into my notebook, run it, copy the output/errors back into the chat, ask Claude to revise, and repeat. It worked, but every iteration meant manually shuttling code and results back and forth between two windows.

With Claude Code as an agent, that loop collapses — it reads and writes files directly in my project, runs its own steps, and only surfaces results when there's something for me to review or approve. Less copy-pasting, more actual delegation.

Next up, I want to try defining a proper workflows/ recipe for recurring research tasks like this one, so I don't have to write the full prompt out each time. Will report back.

Comments

Post a Comment

Popular posts from this blog

How To Implement Dense_Rank() In Excel Sheet

Image
  How To Implement Dense_Rank() In Excel Sheet RANK(), ROW_NUMBER() and DENSE_RANK() are three useful functions in SQL query for ranking purpose. RANK () returns the rank of each row within the partition of a result set. The rank of a row is one plus the number of ranks that come before the row in question. RANK provides the same numeric value for ties For example: 1,2,2,4, 5. The numbers in ranking are not consecutive, there is gap if there are ties. ROW_NUMBER and RANK are similar. ROW_NUMBER numbers all rows sequentially (for example 1, 2, 3, 4, 5). DENSE_RANK() returns the rank of each row within a result set partition, with no gaps in the ranking values. The rank of a specific row is one plus the number of distinct rank values that come before that specific row. By running the T-SQL script in SQL server as below, I got the results that 2 and 4 were skipped due to the ties with RANK(); ROW_NUMBER() gave the consecutive numbers for each row within the r...
Read more

Schema and Security Control in SQL server

Image
dbo is always seen as a SQL server table's schema. dbo stands for database owner. The dbo schema is the default schema of every database. When user created with  CREATE USER   T-SQL command have dbo as their default schema. The dbo schema is owned by the dbo user account.  SQL server also provides other pre-defined schemas that have the same names as the built-in db users and roles: db_accessadmin, db_backupoperator, db_datareader, db_datawriter, db_ddladmin, db_denydatareader, db_denydatawriter, db_owner, db_securityadmin . These exist for backward compatibility. The recommendation is to not use them for user objects.  User can drop the schemas that have the same names as the fixed database roles - unless they're already in use, in which case the drop-command will simply return an error and block the drop of the used schema. For example: IF EXISTS (SELECT * FROM sys.schemas WHERE name = N'db_accessadmin') DROP SCHEMA [db_accessadmin] GO If user drops the...
Read more

My Georgia Tech OMSA Year 2 Fall term Takeaway (Fall25)

By Fall 2025, I felt more grounded in OMSA . The self-doubt from my first year hadn’t completely disappeared, but I was more comfortable with the program’s expectations and pace. This term focused heavily on theoretical depth and statistical rigor , and it pushed me to think more carefully about assumptions, modeling choices, and interpretation. ISYE 6740 – Computational Data Analysis ISYE 6740 is a mathematically intensive course that dives deep into the foundations of machine learning and data analysis. It draws heavily on calculus, linear algebra, probability, statistics , and optimization , and emphasizes understanding why algorithms work, not just how to apply them. Topics such as dimensionality reduction , clustering, classification, and optimization are treated from a theoretical perspective, often involving careful derivations and proofs. Prof. Xie is an exceptional instructor—wise, incredibly sharp, and remarkably clear. She has a rare ability to explain mathematically in...
Read more