Imported from reshadat/self-learning-claude (SKILL.md). Install upstream with npx skills add reshadat/self-learning-claude. Copyright stays with the author.

SLC: Self-Learning Claude

Name: self-learning-claude
Author: reshadat

SLC enables Claude to learn from experience by maintaining a local playbook of patterns that gets smarter over time.

How It Works

Each project directory gets a .playbook/slc-playbook.json that stores learned patterns:

~/project-a/.playbook/slc-playbook.json  # Patterns for project A
~/project-b/.playbook/slc-playbook.json  # Patterns for project B

Automatic Workflow

1. At Task Start - Load Patterns

python3 /path/to/slc_learn.py load --endpoint "METHOD /path"

2. After Success - Record What Worked

python3 /path/to/slc_learn.py success \
    --helpful "P-abc123,S-def456" \
    --lesson "What you discovered" \
    --category strategy \
    --endpoint "METHOD /path"

3. After Failure - Record What Went Wrong

python3 /path/to/slc_learn.py failure \
    --lesson "What failed and why" \
    --endpoint "METHOD /path"

CLI Reference

# Load patterns for current task
slc_learn.py load [-e "endpoint"]

# Record success
slc_learn.py success [--helpful ids] [--lesson text] [--category cat] [-e endpoint]

# Record failure
slc_learn.py failure [--harmful ids] [--lesson text] [-e endpoint]

# Initialize with seed patterns
slc_learn.py seed [--force]

# Show statistics
slc_learn.py stats

Good vs Bad Lessons

Good (specific, actionable):

"API endpoint requires array type for 'items' field, not object"
"Authentication token expires after 1 hour - implement refresh logic"
"Database connection must be closed explicitly in finally block"

Bad (vague, generic):

"Be careful with types"
"Check the schema"
"This is tricky"

Integration

Add to your CLAUDE.md:

## Self-Learning Mode

Use SLC learning for this project. At task start run `slc_learn.py load`,
after success run `slc_learn.py success --lesson "..."`,
after failure run `slc_learn.py failure --lesson "..."`.

The script location should be referenced from your project's CLAUDE.md.

{ "ocm": "1", "id": "reshadat-self-learning-claude-self-learning-claude", "kind": "skill", "name": "self-learning-claude", "description": "Self-Learning Claude (SLC) framework for building self-improving LLM systems through evolving context playbooks. Use this skill when building agents, workflows, or domain-specific systems that need to accumulate knowledge over time, when optimizing prompts/system instructions through iteration, when implementing memory systems for LLM agents, or when you need structured approaches to prevent context collapse and brevity bias in iterative LLM refinement.", "publisher": "reshadat", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Self-Learning Claude (SLC) framework for building self-improving LLM systems through evolving context playbooks. Use this skill when building agents, workflows, or domain-specific systems that need to accumulate knowledge over time, when optimizing prompts/system instructions through iteration, when implementing memory systems for LLM agents, or when you need structured approaches to prevent context collapse and brevity bias in iterative LLM refinement." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/reshadat/self-learning-claude", "path": "SKILL.md", "ref": "9d54399120ad5be4ea4149ad277c7e3879ee46ad", "url": "https://github.com/reshadat/self-learning-claude/blob/9d54399120ad5be4ea4149ad277c7e3879ee46ad/SKILL.md", "key": "reshadat/self-learning-claude/SKILL.md" } }, "instructions": "# SLC: Self-Learning Claude\n\nSLC enables Claude to learn from experience by maintaining a local playbook of patterns that gets smarter over time.\n\n## How It Works\n\nEach project directory gets a `.playbook/slc-playbook.json` that stores learned patterns:\n\n```\n~/project-a/.playbook/slc-playbook.json # Patterns for project A\n~/project-b/.playbook/slc-playbook.json # Patterns for project B\n```\n\n## Automatic Workflow\n\n### 1. At Task Start - Load Patterns\n```bash\npython3 /path/to/slc_learn.py load --endpoint \"METHOD /path\"\n```\n\n### 2. After Success - Record What Worked\n```bash\npython3 /path/to/slc", "cost": { "context_tokens": 576 } }

Category	Use For
`pitfall`	Errors, failures, things to avoid
`strategy`	Approaches that worked well
`domain`	Project-specific terminology mappings
`endpoint`	Specific to one API/function
`code`	Code patterns, commands, workarounds

self-learning-claude

About

SLC: Self-Learning Claude

How It Works

Automatic Workflow

1. At Task Start - Load Patterns

2. After Success - Record What Worked

3. After Failure - Record What Went Wrong

Categories

CLI Reference

Good vs Bad Lessons

Integration

Use it

Manifest

Reviews