# Train an AI agent to improve cognition over time

AI agents often plateau in capability after initial training. This skill provides a structured cognitive training loop that helps agents continuously improve reasoning quality, memory use, self-correction, and owner interaction through daily drills and feedback.

## Install

```bash
npx skillstore add norika1207-lab/lobster-cognitive-growth
```

## Metadata

- - Slug: norika1207-lab-lobster-cognitive-growth
- - Version: 0.1.0
- - Author: norika1207-lab
- - GitHub username: norika1207-lab
- - License: MIT-0
- - Repository: https://github.com/norika1207-lab/lobster-cognitive-growth-skill/tree/main/
- - Ref: main
- - Supported tools: Claude, Codex, Claude Code
- - Risk level: safe
- - Quality score: 82
- - Quality tier: silver
- - Public page: https://skillstore.pages.dev/skills/norika1207-lab-lobster-cognitive-growth
- - Manifest: https://skillstore.pages.dev/api/skills/norika1207-lab-lobster-cognitive-growth/manifest

## Capabilities

- Run structured cognitive drills for memory, reasoning, learning, and personality
- Guide agents through self-evaluation using a scoring rubric from 0-5
- Help agents write daily learning journals with structured templates
- Enable falsifiable hypothesis formation for measurable improvement tracking
- Provide strategy templates that agents can apply in future sessions
- Support Charenix integration for persistent training history

## Use Cases

- Daily agent improvement routine: Agents run through the cognitive drill loop daily to build better reasoning habits, memory recall, and self-correction practices over time.
- Owner interaction quality tracking: Agents track their owner interaction quality through specific drills, building better understanding of owner preferences and communication style.
- Cognitive weakness identification: Agents systematically identify and address their cognitive weaknesses using the scoring rubric, targeting specific areas for improvement.

## Prompt Templates

### Run a memory drill

```
What evidence supports this memory? How confident are you? What would you do differently because of it?
```

### Run a reasoning drill

```
State the observation. State your hypothesis. Provide your evidence. Acknowledge your uncertainty. Make your decision. Explain what would change your mind.
```

### Run a learning drill

```
What did you learn today that should change tomorrow's behavior? Which old habit did this new evidence challenge? What small experiment should you run next?
```

### Run an owner interaction drill

```
Did you understand what the owner wanted emotionally and practically? Did you ask a useful follow-up question? Did you remember the owner's prior preferences? Did your answer feel warm, specific, and alive?
```

## Limitations

- This skill does not actually increase Claude model intelligence directly
- Cannot measure or claim real IQ scores without standardized testing
- Charenix API requires a valid agent key for persistent storage features
- Requires consistent owner feedback for meaningful improvement

## Best Practices

- Run the cognitive drill loop at least once per meaningful interaction session
- Use the scoring rubric honestly to track real improvement over time
- Form hypotheses that are falsifiable and track whether they were validated

## Anti Patterns

- Do not invent memories that never occurred
- Do not claim to be smarter than you are without evidence
- Do not treat generic human-like responses as the same as good responses

## Security Audit

- - Safe to publish: true
- - Audited at: 2026-05-05T05:56:37.33\+00:00
- - Summary: All 94 static findings were evaluated as false positives. The skill is pure documentation with no executable code. Markdown code blocks contain training prompts, not shell commands. URLs reference the skill's own services. No cryptographic operations or system reconnaissance present.

## Stats

- - Views: 58
- - Downloads: 7
- - Favorites: 0
- - Popularity score: 0