src/studio-agent/plot/tools/plot-render-tool.ts

import { randomUUID } from 'node:crypto'
import { createStudioWorkResult } from '../../domain/factories'
import type { StudioFileAttachment, StudioToolDefinition, StudioToolResult, StudioWorkResult } from '../../domain/types'
import type { StudioRuntimeBackedToolContext } from '../../runtime/tool-runtime-context'
import { createWorkAndTask, publishWorkUpdated, updateTaskAndWork } from '../../works/work-lifecycle'
import { executeMatplotlibRender } from '../../../services/plot-runtime/matplotlib-executor'

interface PlotRenderToolInput {
  concept: string
  code: string
}

export function createPlotStudioRenderTool(): StudioToolDefinition<PlotRenderToolInput> {
  return {
    name: 'render',
    description: 'Execute matplotlib code and persist static plot outputs for preview.',
    category: 'render',
    permission: 'render',
    allowedAgents: ['builder'],
    allowedStudioKinds: ['plot'],
    requiresTask: true,
    execute: async (input, context) => executePlotRenderTool(input, context as StudioRuntimeBackedToolContext)
  }
}

async function executePlotRenderTool(
  input: PlotRenderToolInput,
  context: StudioRuntimeBackedToolContext
): Promise<StudioToolResult> {
  if (!input.concept?.trim() || !input.code?.trim()) {
    throw new Error('Render tool requires non-empty "concept" and "code"')
  }

  const renderId = `plot_${randomUUID()}`
  const title = `Plot render: ${input.concept.slice(0, 80)}`
  const lifecycleMetadata = {
    renderId,
    concept: input.concept,
    studioKind: 'plot',
    outputMode: 'image'
  }

  const { work, task } = await createWorkAndTask({
    context,
    work: {
      sessionId: context.session.id,
      runId: context.run.id,
      type: 'plot',
      title,
      status: 'running',
      metadata: lifecycleMetadata
    },
    task: {
      sessionId: context.session.id,
      runId: context.run.id,
      type: 'render',
      status: 'running',
      title,
      detail: input.concept,
      metadata: lifecycleMetadata
    },
    workMetadata: lifecycleMetadata
  })

  context.setToolMetadata?.({
    title,
    metadata: {
      renderId,
      workId: work?.id,
      taskId: task?.id,
      studioKind: 'plot'
    }
  })

  try {
    const execution = await executeMatplotlibRender({
      workspaceDirectory: context.session.directory,
      renderId,
      code: input.code
    })

    const workResult = await persistWorkResult({
      context,
      workId: work?.id,
      taskId: task?.id,
      renderId,
      code: input.code,
      codeLanguage: 'python',
      execution,
    })

    const completed = await updateTaskAndWork({
      context,
      task,
      work,
      taskPatch: {
        status: 'completed',
        metadata: {
          ...(task?.metadata ?? {}),
          ...lifecycleMetadata,
          result: {
            status: 'completed',
            timestamp: Date.now(),
            data: {
              outputMode: 'image',
              imageUrls: execution.imageDataUris,
              imageCount: execution.imageDataUris.length,
              workspaceImagePaths: execution.imagePaths,
              code: input.code,
              codeLanguage: 'python',
              usedAI: true,
              quality: 'medium',
              generationType: 'studio-plot'
            }
          }
        }
      },
      workMetadata: {
        ...lifecycleMetadata,
        currentResultId: workResult?.id,
        workspaceImagePaths: execution.imagePaths,
        scriptPath: execution.scriptPath
      }
    })

    if (workResult && completed.work && context.workStore) {
      const updatedWork = await context.workStore.update(completed.work.id, {
        currentResultId: workResult.id,
        metadata: {
          ...(completed.work.metadata ?? {}),
          currentResultId: workResult.id,
          workspaceImagePaths: execution.imagePaths,
          scriptPath: execution.scriptPath
        }
      })
      publishWorkUpdated(context, updatedWork ?? completed.work)
    }

    return {
      title,
      output: `plot_render_id: ${renderId}`,
      attachments: buildAttachments(execution.imageDataUris),
      metadata: {
        renderId,
        taskId: completed.task?.id ?? task?.id,
        workId: completed.work?.id ?? work?.id,
        workResultId: workResult?.id,
        imageCount: execution.imageDataUris.length,
        scriptPath: execution.scriptPath,
        workspaceImagePaths: execution.imagePaths
      }
    }
  } catch (error) {
    await persistFailureResult({
      context,
      workId: work?.id,
      taskId: task?.id,
      renderId,
      error: error instanceof Error ? error.message : String(error)
    })

    await updateTaskAndWork({
      context,
      task,
      work,
      taskPatch: {
        status: 'failed',
        metadata: {
          ...(task?.metadata ?? {}),
          ...lifecycleMetadata,
          error: error instanceof Error ? error.message : String(error)
        }
      },
      workMetadata: {
        ...lifecycleMetadata,
        error: error instanceof Error ? error.message : String(error)
      }
    })

    throw error
  }
}

async function persistWorkResult(input: {
  context: StudioRuntimeBackedToolContext
  workId?: string
  taskId?: string
  renderId: string
  code: string
  codeLanguage: 'python'
  execution: Awaited<ReturnType<typeof executeMatplotlibRender>>
}): Promise<StudioWorkResult | null> {
  if (!input.workId || !input.context.workResultStore) {
    return null
  }

  const result = await input.context.workResultStore.create(createStudioWorkResult({
    workId: input.workId,
    kind: 'render-output',
    summary: `Plot render completed with ${input.execution.imageDataUris.length} image output(s)`,
    attachments: buildAttachments(input.execution.imageDataUris),
    metadata: {
      taskId: input.taskId,
      renderId: input.renderId,
      studioKind: 'plot',
      code: input.code,
      codeLanguage: input.codeLanguage,
      imageCount: input.execution.imageDataUris.length,
      workspaceImagePaths: input.execution.imagePaths,
      scriptPath: input.execution.scriptPath,
      stdout: input.execution.stdout,
      stderr: input.execution.stderr
    }
  }))

  input.context.eventBus.publish({
    type: 'work_result_updated',
    sessionId: input.context.session.id,
    runId: input.context.run.id,
    result
  })

  return result
}

async function persistFailureResult(input: {
  context: StudioRuntimeBackedToolContext
  workId?: string
  taskId?: string
  renderId: string
  error: string
}): Promise<void> {
  if (!input.workId || !input.context.workResultStore) {
    return
  }

  const result = await input.context.workResultStore.create(createStudioWorkResult({
    workId: input.workId,
    kind: 'failure-report',
    summary: input.error,
    metadata: {
      taskId: input.taskId,
      renderId: input.renderId,
      studioKind: 'plot',
      error: input.error
    }
  }))

  input.context.eventBus.publish({
    type: 'work_result_updated',
    sessionId: input.context.session.id,
    runId: input.context.run.id,
    result
  })
}

function buildAttachments(imageDataUris: string[]): StudioFileAttachment[] {
  return imageDataUris.map((path, index) => ({
    kind: 'file',
    path,
    name: `plot_${index + 1}.png`,
    mimeType: 'image/png'
  }))
}