Are you an LLM? Read llms.txt for a summary of the docs, or llms-full.txt for the full context.
Skip to content

Conversation Skill

Classifies user intent and manages conversation context for the /agentful command.

What It Provides

Intent Classification Knowledge
  • Patterns for recognizing feature requests, bug reports, status queries
  • Methods for extracting information from natural language
  • Ambiguity detection and resolution strategies
Context Management Procedures
  • How to track conversation state across messages
  • Pronoun resolution techniques
  • Session continuity patterns

Intent Categories

The skill defines these intent types:

IntentPatternsAction
feature_request"add", "create", "implement"Delegate to orchestrator
bugfix_request"fix", "broken", "not working"Start bugfix workflow
status_query"status", "progress", "how's it going"Read state, report progress
validation_request"validate", "check quality", "run tests"Trigger validation
continue_request"continue", "next", "keep going"Resume last task
product_query"what are we building", "show product"Display product spec

How Agents Use This

The /agentful command uses this skill to:

  1. Read user input
  2. Match against intent patterns
  3. Extract relevant entities (feature names, requirements)
  4. Check conversation history for context
  5. Classify intent and route to appropriate handler

Context Structure

Conversation history stored in .agentful/conversation-history.json:

{
  "version": "1.0",
  "session_id": "uuid",
  "messages": [
    {
      "user_input": "Add user authentication",
      "classified_intent": "feature_request",
      "extracted_entities": {
        "feature": "user authentication"
      },
      "timestamp": "2026-01-19T02:00:00Z"
    }
  ],
  "context": {
    "last_intent": "feature_request",
    "last_feature": "user authentication",
    "conversation_state": "active"
  }
}

Examples

Feature Request

Input: "Add user profiles with avatar upload"
 
Classification:
  intent: feature_request
  entities: {
    feature: "user profiles",
    requirements: ["avatar upload"]
  }
 
Action: Delegate to orchestrator agent

Status Query

Input: "How's authentication going?"
 
Classification:
  intent: status_query
  target: "authentication"
 
Action: Read .agentful/state.json, report progress

Context Resolution

Message 1: "Add authentication"
Message 2: "Make it use NextAuth"
 
Context resolution:
  "it" → "authentication" (from last message)
 
Action: Update authentication approach to use NextAuth

Implementation Details

Location: .claude/skills/conversation/SKILL.md

Model: Sonnet (optimized for language understanding)

Tools Used:
  • Read (conversation history, state files)
  • Write (update conversation history)
  • Glob (find relevant files)
  • Grep (search for context)

Common Patterns

Ambiguity Detection

If user input matches multiple features:
1. List all matches
2. Ask user to clarify
3. Wait for specific selection

Mind Change Handling

If new intent contradicts previous:
1. Detect contradiction
2. Confirm with user
3. Update context to new direction

Session Resumption

If >30 minutes since last message:
1. Summarize last state
2. Ask if user wants to continue
3. Provide context refresh

Usage by Commands

/agentful
  • Primary consumer of this skill
  • Uses for every user interaction
  • Routes based on classified intent
/agentful-status
  • Uses context to understand what "it" refers to
  • Determines what progress to show

Extending the Skill

Add custom intents by editing .claude/skills/conversation/SKILL.md:

## Custom Intent: deployment_request
 
Patterns:
- "deploy to [environment]"
- "ship [feature]"
- "release to production"
 
Action:
1. Extract environment (staging, production)
2. Validate quality gates
3. Delegate to deployment workflow

See Also