DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to understand a full agent loop in one afternoon with Smol

Run Smol, a deliberately tiny Python agent with one shell tool, against any Responses API endpoint, then read its code with the README's questions.

Short answer: Clone the repo and read Smol.py; then run a small local proxy that holds your API key or session and exposes a Responses API endpoint; run python3 smol.py http://127.0.0.1:8787/v1/responses. All steps.

Credentials. Smol adds no credentials; whatever proxy you put in front holds your key or session.

Who it is forA developer or student who uses coding agents and wants to understand what actually happens inside one.

The problemProduction agents are huge, so it is hard to see the core loop, why context matters and where failures come from.

The outcomeA working agent you have read line by line, including retries, context compaction and shell output truncation, and answers to why each choice was made.

What you need first

  • Python 3 or Go
  • An OpenAI Responses API compatible endpoint, such as a local proxy that adds your key

Steps

  1. Clone the repo and read Smol.py
  2. Run a small local proxy that holds your API key or session and exposes a Responses API endpoint
  3. Run python3 smol.py http://127.0.0.1:8787/v1/responses
  4. Type a prompt at the > prompt; empty input exits
  5. Paste the README prompt into your favourite agent: Read the code of Smol.py and answer all questions from the README
  6. Work through the questions on compaction, truncation and why there is no system prompt

Prompt for your agent

Read the code of smol.py from smol-env/smol and answer all questions from the README: what it does, why it has one tool, how compaction keeps a stable prefix and what that does to caching and cost, and its failure modes. Then explain it line by line.

Caveat. Model requested commands run directly through sh, so use a disposable environment.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for smol-env/smol and "Understand a full agent loop in one afternoon"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer