A learning partner skill for AI coding agents, grounded in neuroscience and philosophy.
Install • Actions • Examples • Contributing • Changelog
When AI gives you an answer, it feels like you've learned. It doesn't mean you have.
Researchers call this the illusion of competence: the fluency of reading a well-structured answer is mistaken for actual understanding. Students who re-read notes feel more confident than students who test themselves — but perform significantly worse.
agentic-learning makes your AI assistant resist this pattern. Instead of handing over answers, it guides you through four neuroscience-backed techniques that build real, durable understanding — while you build real software.
| Action | Trigger | What it does |
|---|---|---|
learn |
@agentic-learning learn <topic> |
You explain first; agent corrects and fills the gaps |
quiz |
@agentic-learning quiz |
Retrieval practice on your current file or topic |
reflect |
@agentic-learning reflect |
3-part structured reflection: learned / goal / gap |
space |
@agentic-learning space |
Schedules concepts to revisit; writes docs/revisit.md |
brainstorm |
@agentic-learning brainstorm <idea> |
Design dialogue; writes docs/brainstorm/YYYY-MM-DD-<topic>.md |
explain-first |
@agentic-learning explain-first |
You narrate the code before the agent says anything |
struggle |
@agentic-learning struggle <task> |
3-hint ladder; reveals only when you ask |
either-or |
@agentic-learning either-or <decision> |
Decision journal; appends to docs/decisions/ |
explain |
@agentic-learning explain |
Reads the project; writes comprehension log to docs/project-knowledge.md |
interleave |
@agentic-learning interleave |
Mixed retrieval across multiple topics — the harder, stickier quiz |
cognitive-load |
@agentic-learning cognitive-load <topic> |
Decomposes an overwhelming concept into working-memory-sized steps |
Full walkthroughs with realistic conversations for every action: EXAMPLES.md
Based on 7 neuroscience-backed techniques — retrieval, spacing, generation, reflection, interleaving, cognitive load management, and metacognition. See references/learning-science.md for primary sources including Roediger & Karpicke (2006), Rohrer & Taylor (2007), Sweller (1988), and Lakhani (2021).
The struggle action uses a 4-level hint system that keeps you in productive struggle as long as possible.
Say more hints to advance, or show me to skip to the full solution at any point.
npx skills add FavioVazquez/agentic-learningInstalls to the current workspace via the skills CLI. Also available at skills.sh/FavioVazquez/agentic-learning.
Works with any AgentSkills-compatible agent. Installs to .agents/skills/ (workspace) — the standard path all agents read from.
# Workspace (current project only)
curl -fsSL https://raw.githubusercontent.com/FavioVazquez/agentic-learning/main/install.sh | bash
# Global (all projects) — use --agent to target a specific agent's home directory
curl -fsSL https://raw.githubusercontent.com/FavioVazquez/agentic-learning/main/install.sh | bash -s -- --global --agent claude # Claude Code
curl -fsSL https://raw.githubusercontent.com/FavioVazquez/agentic-learning/main/install.sh | bash -s -- --global --agent windsurf # Windsurf
curl -fsSL https://raw.githubusercontent.com/FavioVazquez/agentic-learning/main/install.sh | bash -s -- --global # universal fallback
# Uninstall
curl -fsSL https://raw.githubusercontent.com/FavioVazquez/agentic-learning/main/install.sh | bash -s -- --uninstall# Workspace (all agents)
git clone --depth 1 https://github.com/FavioVazquez/agentic-learning .agents/skills/agentic-learning
# Global — Claude Code
git clone --depth 1 https://github.com/FavioVazquez/agentic-learning ~/.claude/skills/agentic-learning
# Global — Windsurf
git clone --depth 1 https://github.com/FavioVazquez/agentic-learning ~/.codeium/windsurf/skills/agentic-learningWorks with any AgentSkills-compatible agent, including Cursor, Claude Code, GitHub Copilot, Gemini CLI, Amp, Warp, Cline, Codex, and Windsurf.
When you use this skill, it creates or appends to files in your project — never in the skill directory itself.
docs/
├── brainstorm/
│ └── YYYY-MM-DD-<topic>.md # brainstorm design docs
├── decisions/
│ └── YYYY-MM-DD-decisions.md # either-or decision journal
├── project-knowledge.md # explain: growing comprehension log
└── revisit.md # space: revisit reminders
These files are yours — commit them, use them as documentation, or review them during retrospectives.
Learning requires productive struggle — the mental effort that builds real understanding. Nine techniques are proven to work:
| Technique | What it means | Action that uses it |
|---|---|---|
| Retrieval | Recalling from memory beats re-reading | learn, quiz, explain-first |
| Spacing | Returning to topics over time beats cramming | space |
| Generation | Producing an answer (even wrong) builds stronger traces | learn, struggle |
| Reflection | Structured feedback on goals and gaps improves outcomes | reflect, either-or |
| Interleaving | Mixing topics beats studying one topic to exhaustion | interleave |
| Cognitive load | Working memory is limited — decompose before you explain | cognitive-load |
| Metacognition | Awareness of how you're learning, not just what | reflect, learn |
| Oracy | Articulating ideas in your own words surfaces gaps retrieval alone doesn't | explain-first |
| Formative feedback | Feedback tied to process and goal, not just correct/incorrect | learn, quiz, struggle |
See references/learning-science.md for all 9 techniques and primary sources, including Inadequate by Priya Lakhani (2021), Hattie & Timperley (2007), and Dweck (2006).
Kierkegaard's Either/Or (1843) argues that every significant choice defines who we are — not just the option chosen, but the act of choosing consciously. Applied to building software with AI agents: every architectural decision, every tradeoff, every "ship it vs. refactor" moment is worth recording. The act of articulating the paths and the rationale is itself learning.
Contributions are welcome. Read CONTRIBUTING.md for how to add actions, improve existing ones, and what "correct behavior" means for this skill.
- TESTING.md — behavior verification checklist for all 9 actions
- CHANGELOG.md — version history
- CODE_OF_CONDUCT.md — community standards
MIT — see LICENSE
Built by Favio Vázquez.
Mental effort is not a flaw in the process. It is the process.