app/services/intelligence.py

import re
from copy import deepcopy
from typing import Any

from app.models.workflow import (
    ContractResponse,
    DependencyImpactResponse,
    DocumentationResponse,
    EnvironmentPromotionResponse,
    IntentDriftResponse,
    LineageField,
    LineageResponse,
    QualityResponse,
    ReleasePlanRequest,
    ReleasePlanResponse,
    RoiResponse,
    SelfHealResponse,
    ValidationIssue,
    WebhookInspectResponse,
    WorkflowDocument,
    WorkflowPackageRequest,
    WorkflowPackageResponse,
)
from app.services.operations import estimate_cost
from app.services.optimizer import optimizer
from app.services.validation import validator


FIELD_REF = re.compile(r"\$json\.([A-Za-z_][\w.]*)")
ENV_REF = re.compile(r"\$\{ENV\.([A-Z][A-Z0-9_]*)\}")
WORD = re.compile(r"[a-z][a-z0-9_-]{2,}")
SECRET_NAMES = {"password", "secret", "token", "api_key", "apikey", "authorization"}
PERSONAL_NAMES = {"email", "phone", "address", "name", "ssn", "dob"}
FINANCIAL_NAMES = {"amount", "invoice", "card", "iban", "account", "price"}
STOP_WORDS = {"that", "with", "from", "this", "then", "into", "workflow", "create", "build"}


def _walk(value: Any, prefix: str = ""):
    if isinstance(value, dict):
        for key, nested in value.items():
            path = f"{prefix}.{key}" if prefix else str(key)
            yield path, nested
            yield from _walk(nested, path)
    elif isinstance(value, list):
        for index, nested in enumerate(value):
            yield from _walk(nested, f"{prefix}[{index}]")


def _classification(field: str) -> str:
    parts = {part.lower() for part in re.split(r"[.\[\]_-]+", field) if part}
    if parts & SECRET_NAMES:
        return "secret"
    if parts & FINANCIAL_NAMES:
        return "financial"
    if parts & PERSONAL_NAMES:
        return "personal"
    return "internal"


def lineage(workflow: WorkflowDocument) -> LineageResponse:
    dependencies: dict[str, list[str]] = {node.id: [] for node in workflow.nodes}
    fields: dict[str, dict[str, set[str] | str]] = {}
    labels = {node.id: node.data.label for node in workflow.nodes}
    for edge in workflow.edges:
        if edge.target in dependencies and edge.source in labels:
            dependencies[edge.target].append(labels[edge.source])
    for node in workflow.nodes:
        for path, value in _walk(node.data.parameters):
            candidates = {path}
            if isinstance(value, str):
                candidates.update(FIELD_REF.findall(value))
            for field in candidates:
                record = fields.setdefault(
                    field,
                    {"sources": set(), "consumers": set(), "classification": _classification(field)},
                )
                if field == path:
                    record["sources"].add(node.data.label)  # type: ignore[union-attr]
                else:
                    record["consumers"].add(node.data.label)  # type: ignore[union-attr]
    result = [
        LineageField(
            field=field,
            source_nodes=sorted(record["sources"]),  # type: ignore[arg-type]
            consumer_nodes=sorted(record["consumers"]),  # type: ignore[arg-type]
            classification=record["classification"],  # type: ignore[arg-type]
        )
        for field, record in sorted(fields.items())
    ]
    return LineageResponse(
        fields=result,
        node_dependencies=dependencies,
        sensitive_paths=[field.field for field in result if field.classification in {"secret", "personal", "financial"}],
    )


def _type_name(value: Any) -> str:
    if value is None:
        return "null"
    if isinstance(value, bool):
        return "boolean"
    if isinstance(value, (int, float)):
        return "number"
    if isinstance(value, dict):
        return "object"
    if isinstance(value, list):
        return "array"
    return "string"


def contract_check(sample: dict[str, Any], expected: dict[str, str]) -> ContractResponse:
    inferred = {path: _type_name(value) for path, value in _walk(sample) if not isinstance(value, (dict, list))}
    violations = []
    for path, expected_type in expected.items():
        actual = inferred.get(path)
        if actual is None:
            violations.append(f"Missing required field: {path}")
        elif actual != expected_type:
            violations.append(f"{path} expected {expected_type}, received {actual}")
    return ContractResponse(valid=not violations, inferred_schema=inferred, violations=violations)


def quality(workflow: WorkflowDocument) -> QualityResponse:
    validation = validator.validate(workflow)
    findings = list(validation.issues)
    lineage_result = lineage(workflow)
    code_nodes = [node for node in workflow.nodes if node.data.type == "n8n-nodes-base.code"]
    if code_nodes:
        findings.append(ValidationIssue(code="review_code", severity="warning", message="Code nodes require review and sandboxed testing."))
    if lineage_result.sensitive_paths:
        findings.append(ValidationIssue(code="sensitive_data", severity="warning", message="Sensitive fields require destination and retention review."))
    retry_nodes = sum("retry" in str(node.data.parameters).lower() for node in workflow.nodes)
    scores = {
        "reliability": max(0, validation.score - (10 if retry_nodes == 0 else 0)),
        "maintainability": max(0, 100 - len(code_nodes) * 12 - max(0, len(workflow.nodes) - 30)),
        "security": max(0, 100 - len(lineage_result.sensitive_paths) * 4 - len(code_nodes) * 8),
        "cost": max(0, 100 - sum("langchain" in node.data.type for node in workflow.nodes) * 7),
        "observability": 90 if workflow.settings.get("saveExecutionProgress") else 55,
    }
    return QualityResponse(overall=round(sum(scores.values()) / len(scores)), scores=scores, findings=findings)


def intent_drift(workflow: WorkflowDocument, requirement: str) -> IntentDriftResponse:
    terms = {term for term in WORD.findall(requirement.lower()) if term not in STOP_WORDS}
    workflow_text = " ".join(
        f"{node.data.label} {node.data.type} {node.data.subtitle or ''}" for node in workflow.nodes
    ).lower()
    covered = sorted(term for term in terms if term in workflow_text)
    missing = sorted(terms - set(covered))
    return IntentDriftResponse(
        alignment_score=round(len(covered) / max(1, len(terms)) * 100),
        covered_terms=covered,
        missing_terms=missing,
    )


def promote(workflow: WorkflowDocument, environment: str, values: dict[str, Any]) -> EnvironmentPromotionResponse:
    promoted = deepcopy(workflow)
    replacements = 0
    unresolved: set[str] = set()

    def replace(value: Any) -> Any:
        nonlocal replacements
        if isinstance(value, str):
            def substitute(match: re.Match[str]) -> str:
                nonlocal replacements
                key = match.group(1)
                if key not in values:
                    unresolved.add(key)
                    return match.group(0)
                replacements += 1
                return str(values[key])
            return ENV_REF.sub(substitute, value)
        if isinstance(value, dict):
            return {key: replace(nested) for key, nested in value.items()}
        if isinstance(value, list):
            return [replace(nested) for nested in value]
        return value

    for node in promoted.nodes:
        node.data.parameters = replace(node.data.parameters)
    promoted.meta.model_extra["environment"] = environment
    return EnvironmentPromotionResponse(
        workflow=promoted, environment=environment, replacements=replacements, unresolved=sorted(unresolved)
    )


def release_plan(request: ReleasePlanRequest) -> ReleasePlanResponse:
    validation = validator.validate(request.workflow)
    blocked = not validation.valid
    strategy_steps = {
        "shadow": ["Mirror sanitized input to the candidate version", "Suppress side-effect nodes", "Compare schemas, outputs, and latency"],
        "canary": [f"Route {request.traffic_percentage}% of eligible traffic to the candidate", "Monitor success rate and latency", "Increase traffic only after approval"],
        "synthetic": ["Schedule representative test inputs", "Verify credentials and contracts", "Alert on consecutive failures"],
    }
    return ReleasePlanResponse(
        strategy=request.strategy,
        status="blocked" if blocked else "draft",
        steps=strategy_steps[request.strategy],
        rollback_conditions=[
            f"Error rate exceeds {request.max_error_rate:.1%}",
            f"Success rate drops below {request.success_threshold:.1%}",
            "Output contract or sensitive-data policy fails",
        ],
        warnings=[issue.message for issue in validation.issues if issue.severity == "error"],
    )


def package_workflow(request: WorkflowPackageRequest) -> WorkflowPackageResponse:
    return WorkflowPackageResponse(
        manifest={
            "format": "flowforge.workflow-package/v1",
            "name": request.workflow.name,
            "workflowVersion": request.workflow.meta.version or 1,
            "n8nCompatible": True,
            "contents": ["workflow", "tests", "contracts", "environments"],
        },
        workflow=request.workflow,
        tests=request.tests,
        contracts=request.contracts,
        environments=request.environments,
    )


def documentation(workflow: WorkflowDocument) -> DocumentationResponse:
    nodes = "\n".join(f"- **{node.data.label}** (`{node.data.type}`): {node.data.subtitle or 'Configured action'}" for node in workflow.nodes)
    connections = "\n".join(
        f"- {next((n.data.label for n in workflow.nodes if n.id == edge.source), edge.source)} -> "
        f"{next((n.data.label for n in workflow.nodes if n.id == edge.target), edge.target)}"
        for edge in workflow.edges
    )
    return DocumentationResponse(markdown=f"# {workflow.name}\n\n{workflow.meta.description or 'n8n workflow'}\n\n## Nodes\n\n{nodes}\n\n## Connections\n\n{connections or '- None'}\n")


def roi(workflow: WorkflowDocument, executions: int, minutes_saved: float, hourly_rate: float, sla_minutes: float) -> RoiResponse:
    hours = executions * minutes_saved / 60
    cost = estimate_cost(workflow, executions).estimated_monthly_usd
    estimated_duration = max(1, len(workflow.nodes) * 250)
    return RoiResponse(
        hours_saved=round(hours, 2), labor_value_usd=round(hours * hourly_rate, 2),
        estimated_operating_cost_usd=cost, net_value_usd=round(hours * hourly_rate - cost, 2),
        estimated_duration_ms=estimated_duration,
        sla_headroom_percent=round(max(0, 1 - estimated_duration / (sla_minutes * 60_000)) * 100, 2),
    )


def inspect_webhook(payload: dict[str, Any], redact: bool) -> WebhookInspectResponse:
    result = deepcopy(payload)
    schema = {path: _type_name(value) for path, value in _walk(payload) if not isinstance(value, (dict, list))}
    redacted = []
    if redact:
        def scrub(value: Any, prefix: str = "") -> None:
            if not isinstance(value, dict):
                return
            for key, nested in value.items():
                path = f"{prefix}.{key}" if prefix else key
                if _classification(path) in {"secret", "personal", "financial"}:
                    value[key] = "[REDACTED]"
                    redacted.append(path)
                else:
                    scrub(nested, path)
        scrub(result)
    return WebhookInspectResponse(payload=result, schema_map=schema, redacted_fields=redacted)


def dependency_impact(workflow: WorkflowDocument, dependency: str) -> DependencyImpactResponse:
    query = dependency.lower()
    affected = [node for node in workflow.nodes if query in f"{node.data.type} {node.data.label} {node.data.parameters}".lower()]
    affected_ids = {node.id for node in affected}
    downstream_ids: set[str] = set()
    frontier = list(affected_ids)
    while frontier:
        source = frontier.pop()
        for edge in workflow.edges:
            if edge.source == source and edge.target not in downstream_ids:
                downstream_ids.add(edge.target)
                frontier.append(edge.target)
    labels = {node.id: node.data.label for node in workflow.nodes}
    count = len(affected_ids | downstream_ids)
    severity = "none" if count == 0 else "low" if count == 1 else "medium" if count < 5 else "high"
    return DependencyImpactResponse(
        affected_nodes=[node.data.label for node in affected],
        downstream_nodes=[labels[node_id] for node_id in downstream_ids if node_id in labels],
        severity=severity,
    )


def self_heal(workflow: WorkflowDocument, errors: list[str]) -> SelfHealResponse:
    before = quality(workflow)
    optimized = optimizer.optimize(workflow)
    after = quality(optimized.workflow)
    changes = [suggestion.title for suggestion in optimized.suggestions]
    if errors:
        changes.append("Attached execution errors as approval evidence")
    return SelfHealResponse(
        proposed_workflow=optimized.workflow,
        changes=changes,
        quality_before=before.overall,
        quality_after=after.overall,
    )