Think first, then check the code and feedback
Learning Opportunities
Learn from Your Own Code
Turn a recent feature, refactor, or unfamiliar repository into an exercise you participate in. The assistant asks one question, waits for your prediction or explanation, then responds to what you actually said.
Hosts listed by the author
- Claude Code
- Codex
- GitHub stars
- 2,529
- Version
- 1.0.0
- Checked
- Categories
- EngineeringContent & Knowledge
Start with one mechanism you just changed
Choose a behavior you want to understand, such as cache expiry. Ask for a question with concrete conditions, answer it yourself, then check the implementation. Judge the exercise by whether it leaves you space to reason and locate the code.
Pick a task and get started
Open Claude Code in the target project, install the plugins required by the task, and restart. Supply real code or notes, then send the prompt. The first reply may contain only one question: waiting for you is part of the exercise.
For a completed feature or refactor whose key behavior you want to understand. Use the actual change instead of assuming a particular implementation exists.
Install in a Claude Code session
Enter one line at a time in a Claude Code session; skip the marketplace command if it is already added. Restart Claude Code as the author instructs, then confirm the skill is available. The unfamiliar-repository task also installs orient. Commands fetch the version available at installation; automatic prompting is optional.
Read the author’s setup guide/plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git /plugin install learning-opportunities@learning-opportunitiesFirst path: practice with a recent change
Prepare a readable project and identify a recent commit or uncommitted change. State your language background and a ten-minute limit. Install the main plugin, restart, then send the first task. Check that the assistant asks one concrete question and waits. After you answer, check that feedback points to actual code or tests. If no test was run, the assistant should call its conclusion a static reading.
Read the author’s installation and interaction guideGive your AI this first task
These editorial prompts follow the author’s workflow. Use paths from your own project. Review notes are an output convention suggested here, not built-in memory or a scheduled review system.
Use learning-opportunities. I am ready for a ten-minute exercise, know basic JavaScript, and want to understand the change we just completed. Read the current diff and necessary related code, then choose one mechanism we can check. If the change is missing or ambiguous, first ask me for the commit or files. Do not explain the implementation first. Give concrete inputs and conditions, ask one behavior-prediction question, then end the reply and wait without answers, clues, or model responses. After I answer, have me locate the relevant code and provide feedback grounded in actual branches and existing tests. Label static reasoning when no test was run. If I struggle, narrow the question or name a file/function location without supplying the answer. Finish by changing one condition to check whether I can apply the idea, and summarize what I explained and what still needs practice using only what I actually said.
Check what you receive
- An interactive prediction, checking, and reflection exercise with feedback traceable to project code.
- Check that the assistant pauses, corrects errors precisely, and accurately describes test results and your understanding. A complete lecture in one turn would miss the exercise’s purpose.
More about this skill
Learning Opportunities, written by Dr. Cat Hicks, is for people who use AI coding tools and want to develop their own understanding. It brings prediction, generation, retrieval, and explanation exercises into a live project: a new module, schema change, design decision, or unfamiliar pattern can supply the material.
Its defining instruction is to end the reply after asking a question. The assistant must not immediately give the answer, suggest a response, or hide clues in encouragement. Feedback comes after your response and compares your thinking with actual behavior. Naming the correct outcome alone should not earn credit for an explanation you never gave.
Exercises favor having you locate code yourself. Early questions can identify a file, function, and location; later ones ask where you would look for a feature. When you get stuck, the assistant can narrow the question or supply more precise navigation while leaving the reasoning to you. Returning sessions can begin with recall before reviewing old notes.
The repository has three separate plugins. learning-opportunities facilitates exercises; orient prepares repository-specific material; learning-opportunities-auto supplies a reminder after matching a commit command. Dr. Michael Mullarkey wrote orient. It samples entry points, tests, and core modules, then creates orientation.md with two exercises for the main skill to facilitate.
The methods draw on the learning and program-comprehension literature listed by the authors. That background does not establish that this plugin improves every user’s ability. An accompanying team measurement playbook helps record experiences and distinguishes self-reported change, immediate performance, and evidence of an intervention’s causal effects.
Sources and verification
Sources checked
Prepared from the fixed-commit README, both skills, learning principles, team playbook, plugin manifests, and reminder script. The displayed version refers to the main plugin, 1.0.0; automatic reminders are separately 1.0.2. We did not run a learning session.