07.01 · JIRA to Pull Request — Deep Dive (Case 1)

graph TD
    A[JIRA ticket created / assigned] --> B[Webhook → Agent Service]
    B --> C[LangGraph: read_ticket node]
    C --> D[classify_intent: bug or feature?]
    D -->|Bug| E[Bug Fix Subgraph]
    D -->|Feature| F[Feature Impl Subgraph]
    D -->|Ambiguous| G[INTERRUPT: clarify with author]
    E --> H[human_review INTERRUPT]
    F --> H
    H -->|approved| I[create_pr node]
    H -->|rejected| J[revise node]
    J --> H
    I --> K[PR opened + JIRA linked]

graph LR
    A[Read JIRA bug] --> B[Extract: error, stack trace, steps]
    B --> C[RAG: search code index for relevant classes]
    C --> D[ReAct loop: read files, trace call chain]
    D --> E[Identify root cause]
    E --> F[Generate minimal diff]
    F --> G[Generate regression test]
    G --> H[Validate: does diff + test make sense?]
    H -->|Yes| I[Prepare PR metadata]
    H -->|No| D

graph LR
    A[Read JIRA feature] --> B[Extract acceptance criteria]
    B --> C[Decompose into implementation plan]
    C --> D[Identify affected service and files]
    D --> E[Generate implementation per AC]
    E --> F[Generate test per AC]
    F --> G[Check: all ACs have tests?]
    G -->|Yes| H[Prepare PR metadata]
    G -->|No| E

class JiraAgentState(TypedDict):
    # Input
    ticket_key: str
    ticket_data: dict   # full JIRA API response

    # Classification
    intent: str         # "bug" | "feature" | "ambiguous"
    service_name: str
    confidence: float

    # Retrieval
    retrieved_files: List[str]
    retrieved_context: str

    # Generation
    implementation_plan: List[str]
    code_diff: str
    tests_added: List[str]

    # Review
    review_status: str  # "pending" | "approved" | "rejected"
    reviewer_feedback: str

    # Output
    pr_url: str
    branch_name: str

    # Conversation
    messages: List[BaseMessage]
    iteration_count: int