Analyze Request Skill

Request understanding and routing for DA-Agent

Purpose

Transform vague business requests into structured analysis tasks, score complexity, and route to the appropriate execution mode.

Security Constraints

CRITICAL: Before any file operation, validate paths:

from lib.security_wrapper import SecureFileAccess

security = SecureFileAccess(os.getcwd())
safe_path = security.validate_path(user_provided_path)

Never access:

• Files outside current working directory
• .env files
• Credential files
• .ssh directories

If user provides forbidden path, respond: "Sorry, for security reasons, I can only access files within the current project directory."

Workflow

Phase 1: Structured Dialog

Ask questions ONE AT A TIME to clarify the request. Use multiple choice when possible.

Use question templates from references/question-templates.md.

Required information:

1. Core question - What business question are we answering?
2. Data sources - Which files? (validate with security wrapper)
3. Time range - What period?
4. Dimensions - How to slice the data?
5. Analysis type - What kind of analysis?

Phase 2: Complexity Scoring

Once you have structured requirements, score complexity:

from lib.complexity_scorer import ComplexityScorer

scorer = ComplexityScorer()
result = scorer.score(structured_req)

print(f"Complexity Score: {result.score}/10")
print(f"Suggested Mode: {result.mode}")
print(f"Reasoning: {result.reasoning}")

See references/complexity-scoring-rubric.md for scoring criteria.

Phase 3: User Confirmation

Present the scoring result and ask for confirmation:

Example:

Analysis Task Complexity

Score: 5/10 Suggested Mode: Collaborative Reasoning: Multi-file JOIN (2pts) + Funnel analysis (2pts) + Business rules (1pt)

In collaborative mode:

• I'll generate code drafts for your review
• You'll approve at key decision points
• We'll work together on the analysis

Proceed with this approach?

Phase 4: Route to Execution

Based on confirmed mode:

• auto → Load execute-auto skill and hand off
• collab → (Phase 2: Coming soon)
• assist → (Phase 2: Coming soon)

For auto mode:

# Hand off to execute-auto skill
print("Routing to auto-execution mode...")
# The execute-auto skill will take over from here

Then follow the execute-auto skill workflow.

Phase 5: Knowledge Base Recording

Save the task to knowledge base:

from lib.knowledge_store import KnowledgeStore
import os

kb = KnowledgeStore(Path.home() / ".da-agent" / "knowledge")
task_id = kb.save_task(
    description=original_request,
    structured_req=structured_req,
    complexity_score=result.score,
    execution_mode=result.mode
)

Error Handling

• If security validation fails → Explain why and ask for different path
• If complexity scoring fails → Fall back to rule-based scoring
• If user provides incomplete info → Ask follow-up questions
• If task is too vague → Help decompose into specific questions

Examples

See references/question-templates.md for conversation flow examples.