Skip to content

Reading and preparation

Before class, read the Week 1 row of the calendar and the notes. Recall Python functions, lists, dictionaries and assertions. No machine-learning course is assumed.

Read the five-step run() contract in loop.py and one success, wrong-write and refused-call check in verify_loop.py. Predict the observations before following the live build. The starter is deliberately incomplete.

Primary-source reading

Two open-source coding agents publish the code that builds their instructions under the Apache-2.0 licence: Codex CLI and Qwen Code. Read the pinned files below on GitHub for the prompt comparison. Each project has a reading card that links the files at a fixed commit and records what they contain: one for Codex CLI (tools/codex-base-instructions; as_of: 2026-10-09; Source 1; Source 2; Source 3) and one for Qwen Code (tools/qwen-code-system-prompt; as_of: 2026-10-09; Source 1; Source 2; Source 3; Source 4; Source 5; Source 6; Source 7). Neither set of files is the complete prompt a release sends at run time; your session capture establishes what your run exposed. This page does not rank the two products.

  • Source 1: Codex CLI's base instructions and plan schema. Open the two files linked from the Codex CLI reading card above, pinned at commit 2351d9e1b608e6f9d9a3699b71d7eb39ee41cfa4: the base instructions (codex-rs/protocol/src/prompts/base_instructions/default.md) and the update_plan tool definition (codex-rs/core/src/tools/handlers/plan_spec.rs). Read Task execution, Validating your work and update_plan. Identify a stopping instruction and a required argument.
  • Source 2: Qwen Code's prompt builder and edit tool. In the QwenLM/qwen-code repository at tag v0.25.0 (commit 6788c035698a0ada471c958d1e789e96c6cddd9b), open the prompt builder and the edit tool. Read Software Engineering Tasks and Final Reminder. Separate argument-schema rules, tool validation and requests in prose.

You need not install both agents. Read the files as text, name the enforcing component where visible, and say “cannot tell” where these files do not settle it.

Before the studio

Read the studio steps, then save a prediction before your own agent runs. Practice request indexing with the supplied sample record and viewer. The request questions in the dissection and schema questions in the prompt comparison help you choose citations. Use this week's studio for the task and submission.

After capture, answer the self-check. Bring one observed claim and one inference to discuss. A1 is due this week.