SkillbookToolSet

The SkillbookToolSet provides skill management capabilities for the Skill Learning system, enabling agents to learn, store, and apply knowledge from conversations.

Overview

Key features:

• Skill Management: Add, update, remove, and tag skills

• Section Organization: Categorize skills by type (user_rules, strategies, patterns, workflows)

• Confidence Filtering: Only accept skills above a confidence threshold

• Feedback Tracking: Track helpful/harmful/neutral feedback for quality assessment

• Agent Scoping: Skills can be global or agent-specific

Basic Usage

from pantheon.agent import Agent
from pantheon.toolsets import SkillbookToolSet

# Create skillbook toolset
skillbook_tools = SkillbookToolSet(
    name="skillbook",
    min_confidence=0.7
)

# Create agent and add toolset at runtime
agent = Agent(
    name="skill_manager",
    instructions="You manage and learn from conversations."
)
await agent.toolset(skillbook_tools)

await agent.chat()

Constructor Parameters

Parameter	Type	Description
skillbook	Skillbook | None	Skillbook instance to manage. Auto-creates if None.
min_confidence	float	Minimum confidence threshold for adding skills (default: 0.7)
name	str	Name of the toolset (default: “skillbook”)

Tools Reference

add_skill

Add a new skill to the skillbook.

result = skillbook_tools.add_skill(
    section="strategies",
    content="Use polars lazy evaluation for memory-efficient large file processing",
    description="Polars lazy eval for large files",
    agent_name="data_analyst",
    confidence=0.9
)

Parameters:

• section: Category - “user_rules”, “strategies”, “patterns”, or “workflows”

• content: The skill content (specific and actionable)

• description: Short summary (max 15 words) for long content

• agent_name: Agent scope - “global” for all agents or specific name

• sources: Optional list of temp file paths with examples/documentation

• skill_id: Custom ID (auto-generated if omitted)

• confidence: Confidence level 0.0-1.0 (must exceed min_confidence)

Returns:

{"success": True, "skill_id": "str-a1b2c3d4", "message": "Added skill..."}

update_skill

Update an existing skill’s content or sources.

result = skillbook_tools.update_skill(
    skill_id="str-a1b2c3d4",
    content="Updated skill content here",
    description="Updated description"
)

Parameters:

• skill_id: ID of the skill to update

• content: New content (None to keep unchanged)

• description: New description (None to keep unchanged)

• sources: New source files (replaces existing)

remove_skill

Permanently remove a skill.

result = skillbook_tools.remove_skill(skill_id="str-a1b2c3d4")

tag_skill

Record feedback on a skill’s effectiveness.

result = skillbook_tools.tag_skill(
    skill_id="str-a1b2c3d4",
    tag="helpful"  # or "harmful" or "neutral"
)

Returns:

{
    "success": True,
    "skill_id": "str-a1b2c3d4",
    "stats": "+3/-1/~0",  # helpful/harmful/neutral counts
    "message": "Tagged skill..."
}

list_skills

List and filter skills in the skillbook.

result = skillbook_tools.list_skills(
    section="strategies",
    tag="helpful",
    keyword="polars",
    include_full_content=False
)

Parameters:

• section: Filter by category (None for all)

• tag: Filter by dominant feedback (“helpful”, “harmful”, “neutral”)

• keyword: Case-insensitive search in content

• include_full_content: Return complete content (default: truncated)

Returns:

{
    "total": 5,
    "filters": {"section": "strategies", ...},
    "skills": [
        {
            "id": "str-a1b2c3d4",
            "section": "strategies",
            "content": "Use polars lazy eval...",
            "sources": [],
            "stats": "+3/-0/~1"
        }
    ]
}

compress_trajectory

Compress a conversation memory for analysis.

result = skillbook_tools.compress_trajectory(
    memory_path="/tmp/memory.json",
    output_dir="/tmp/analysis"
)

Returns:

{
    "success": True,
    "trajectory_path": "/tmp/analysis/trajectory_xxx.txt",
    "details_path": "/tmp/memory.json",
    "skill_ids_cited": ["str-xxx", "pat-yyy"]
}

get_skillbook_content

Get formatted skillbook content for injection.

result = skillbook_tools.get_skillbook_content(agent_name="global")

Skill Sections

Skills are organized into four sections:

• user_rules: User preferences and explicit instructions

• strategies: Problem-solving approaches and methodologies

• patterns: Reusable code patterns and templates

• workflows: Multi-step procedures and pipelines

Examples

Learning from a Conversation

# After successful task completion
result = skillbook_tools.add_skill(
    section="strategies",
    content="When processing large CSV files with polars, use scan_csv() "
            "instead of read_csv() for streaming to avoid memory issues",
    agent_name="data_analyst",
    confidence=0.9
)

Reviewing Skills

# Find helpful strategies
result = skillbook_tools.list_skills(
    section="strategies",
    tag="helpful"
)

# Search for specific topics
result = skillbook_tools.list_skills(keyword="polars")

Providing Feedback

# After a skill helped
skillbook_tools.tag_skill(skill_id="str-abc123", tag="helpful")

# After a skill caused issues
skillbook_tools.tag_skill(skill_id="str-xyz789", tag="harmful")

Analyzing Conversations

# Compress conversation for analysis
result = skillbook_tools.compress_trajectory(
    memory_path="/path/to/memory.json"
)

# Check which skills were referenced
print(f"Skills cited: {result['skill_ids_cited']}")

Best Practices

1. Check before adding: Use list_skills to avoid duplicates

2. Be specific: Write actionable, concrete skills

3. Use confidence scores: Only add skills you’re confident about

4. Provide feedback: Tag skills after observing their effectiveness

5. Organize by section: Choose appropriate categories for skills

6. Scope appropriately: Use agent-specific scopes when relevant