Learning Opportunities
Pauses AI-assisted coding to offer science-based 10-15 minute learning exercises, so you build genuine expertise instead of just shipping AI-generated code.
Pure prompt-skill with no scripts, no external side effects, no data collection; explicitly requires user consent before every exercise; orient mode only reads local skill directories and stops with a clear message when files are missing. Deducted for: PRINCIPLES.md and the orient plugin are external links whose integrity isn't verified; the opt-out mechanism is informal (relies on the user declining) with no formal rollback path.
Instructions are self-consistent and well-structured (triggers, suppression conditions, exercise types, hard-stop rules), and the failure path (missing orientation.md) has explicit feedback text. Deducted for: no execution or test evidence in a static review; README itself admits Claude's default behavior may conflict with the 'hard stop' instruction; behavior on abnormal input is undefined; capped at 10 per static calibration.
Trigger conditions (new files, schema changes, refactors, 'why' questions) and suppression conditions (declined this session, 2 completed) are explicit; scope (Claude Code/Codex) is declared. Deducted for: non-fit boundaries only partially specified; orient mode depends on an external orient plugin and third-party repo reachability; no Chinese-language or localization support; audience is English-speaking developers.
Full CC-BY-4.0 license text, clear author identity and background, scholarly source lists, dual install instructions (Claude Code and Codex), and documented optional components. Deducted for: no version number or changelog; maintenance/update path relies on the author's site; parameter convention covers only 'orient' with partial external dependency; known-limitation disclosure incomplete despite good progressive disclosure.
Exercise designs (prediction→observation, generation→comparison, fading scaffolds) align with the cited learning science and are concrete enough for Claude to execute; the README is candid about the cost of interrupting flow. Deducted for: no verified real output samples under static review; value claims rest on literature rather than evidence about this skill; comparative-benefit evidence limited, so not above 7.
PRINCIPLES.md gives an auditable list of peer-reviewed sources and separates findings/mechanisms/applications; the author links public OSF preprints enabling cross-checking of design rationale. Deducted for: no runtime reproducibility evidence (no CI, tests, or output snapshots); the causal link between the literature and this skill's behavior is not independently verified; capped at 5.
- Static source review only; the skill was not executed. Real interaction behavior—especially whether the 'hard stop' instruction can override the model's default to keep generating—is unverified.
- Orient mode depends on an external orient plugin and a locally generated orientation.md; the experience degrades when these are absent, and third-party repos must be reachable.
- No version number or changelog; behavior may change silently between updates—pin the repository revision when deploying.
- No Chinese-language support documented; targeted at English-speaking users; GitHub-hosted resources may be hard to reach from mainland China.
- Publisher is not verified by the FollowSkills registry; identity is treated as unknown and no points were deducted for this alone.
What does this skill do, and when should you use it?
A Claude Code and Codex skill that offers optional 10-15 minute interactive learning exercises after architectural work — new files, schema changes, refactors, or unfamiliar patterns. It targets the "AI productivity trap": fast, fluent AI output can skip the active processing that builds real understanding. Six exercise types (prediction, generation-comparison, tracing, debugging, teach-it-back, retrieval check-ins) are anchored in learning science, with a core mechanism of Claude pausing and waiting for your answer instead of answering its own questions. Created by psychological scientist Dr. Cat Hicks from developer interviews and her peer-reviewed research, licensed CC-BY-4.0.
Detects completion of significant work (new files/modules, schema changes, refactors, unfamiliar patterns) and asks whether you'd like a quick exercise. If accepted, runs one of six formats: Prediction → Observation → Reflection, Generation → Comparison, Trace the Path, Debug This, Teach It Back, or a Retrieval Check-in at the start of returning sessions. Enforces a hard "pause for input" rule: after posing a question it ends its message, offering no hints, example answers, or prompt suggestions. Prefers directing you to locate code in files yourself, with scaffolding that fades as familiarity grows. Invoked with the orient argument, it reads a generated orientation.md and runs repo-specific guided lessons. Suppression rules prevent repeat offers after a decline or two completed exercises per session.
- A developer building a personal project with Claude Code involving unfamiliar languages or architectural patterns who worries they're just accepting AI code without learning
- An engineer onboarding onto a new codebase who wants repo-specific guided lessons via `/learning-opportunities orient`
- A tech lead rolling out agentic coding across a team who wants learning culture to be visible, using the MEASURE-THIS.md pre/post measurement playbook
- A developer on a long-running project who wants session-start retrieval check-ins to reinforce last session's architectural understanding
- A learning-science-curious AI coding user who wants to customize exercise types using the referenced PRINCIPLES.md
What are this skill's strengths and limitations?
- Exercise design rests on solid learning science (generation effect, retrieval practice, spacing, metacognition) with a detailed PRINCIPLES.md explaining the reasoning
- The hard "pause for input" rule directly counters the AI default of answering its own questions, forcing genuine mental effort
- Exercises draw on your own project's code and direct you to locate files rather than reading snippets, building codebase familiarity alongside the concept
- Polite suppression rules (per-session cap, stop after a decline) keep it from becoming an annoying interruption
- Backed by the author's peer-reviewed developer psychology research, plus a team measurement playbook (MEASURE-THIS.md)
- The 10-15 minute pause deliberately breaks the fast agentic flow — pure velocity-focused users may find it frustrating
- The skill depends on the model obeying "stop generating after the question"; the README acknowledges you may need to work around Claude's default to give complete answers, so failure modes exist
- The orient mode requires a separate upfront step of generating an orientation.md file
- No automated test suite in the repo; quantitative evidence comes from the author's research, not a controlled trial of this skill itself
- The auto-trigger hook (learning-opportunities-auto) requires extra configuration on Windows
How do you install this skill?
Claude Code: 1) add the marketplace: /plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git; 2) install: /plugin install learning-opportunities@learning-opportunities; 3) restart Claude Code. Codex: run codex plugin marketplace add https://github.com/DrCatHicks/learning-opportunities.git. (Optional) Linux/macOS users can add learning-opportunities-auto so Claude considers an exercise after each git commit; Windows needs extra setup. (Optional) install the orient plugin with /plugin install orient@learning-opportunities to generate repo orientation files.
How do you use this skill?
After installation, no manual trigger is needed — when Claude detects you've completed new files, schema changes, or refactors, it asks "Would you like to do a quick learning exercise on [topic]? About 10-15 minutes." Accept and answer its questions; Claude deliberately pauses and waits for your input. To learn a new repo, first run /orient in it to generate orientation.md, then call /learning-opportunities orient for guided lessons. Exercises are capped at 2 per session, and one decline silences offers for the session.
How does this skill compare with similar options?
The README explicitly recommends pairing with the same author's Learning-Goal skill: Learning-Goal guides semi-structured learning goal-setting using MCII (Mental Contrasting with Implementation Intentions), while this skill provides in-the-moment exercises during everyday coding — one sets goals, the other delivers daily practice.