@flower-f/afk-pipelines-official-official

Official built-in workflow pipelines for afk-dev.

Pipelines

• feedback-loop — iterative implement → verify → decide → apply loop.
• stage-sequence — ordered stage execution with diff/check propagation.

Usage

import { builtInWorkflowPipelines } from "@flower-f/afk-pipelines-official";
import { createWorkflowPipelineRegistry } from "@flower-f/afk-core";

const registry = createWorkflowPipelineRegistry(builtInWorkflowPipelines);

This package is used by @flower-f/afk-node to assemble the default pipeline registry.

Custom pipelines

WorkflowPipeline is the public extension seam for custom workflow execution. A pipeline implementation owns its workflow contract: stage requirements, pipeline option validation, run behavior, progress events, metrics, and artifacts. Callers pass pipeline.* values generically; the selected pipeline owns their meaning.

import {
  WorkflowPipeline,
  createWorkflowPipelineRegistry,
  type WorkflowPipelineValidationInput,
  type WorkflowRunInput,
  type WorkflowRunResult,
  type WorkflowRuntimeContext,
} from "@flower-f/afk-core";
import { builtInWorkflowPipelines } from "@flower-f/afk-pipelines-official";
import { runAfkWorkflow, runDoctor } from "@flower-f/afk-node";

class DocsGeneratorPipeline extends WorkflowPipeline {
  readonly name = "docs-generator";

  override validate({ workflow }: WorkflowPipelineValidationInput): void {
    if (!workflow.stages.draft) {
      throw new Error(`Workflow '${workflow.name}' must define a draft stage.`);
    }
  }

  override async run(
    input: WorkflowRunInput,
    _runtimes: WorkflowRuntimeContext,
  ): Promise<WorkflowRunResult> {
    const startedAt = new Date().toISOString();
    const completedAt = new Date().toISOString();
    return {
      pipeline: this.name,
      task: input.task,
      startedAt,
      completedAt,
      stopReason: "completed",
      sessions: {},
      metrics: [],
      artifacts: [{ key: "report", kind: "report", filename: "report.json", format: "json" }],
      buildArtifacts: () => [
        {
          key: "report",
          kind: "report",
          filename: "report.json",
          format: "json",
          content: { task: input.task },
        },
      ],
    };
  }
}

const workflowPipelineRegistry = createWorkflowPipelineRegistry([
  ...builtInWorkflowPipelines,
  new DocsGeneratorPipeline(),
]);

await runDoctor({ workflowPipelineRegistry });
await runAfkWorkflow({ task, overrides, workflowPipelineRegistry });

The CLI starts with the built-in registry and extends it with pipelines declared by enabled plugins. Programmatic Node callers can also inject a registry into runDoctor and runAfkWorkflow; plugin-declared pipelines are merged into that same registry before workflow selection, doctor validation, and execution.

Stage-sequence prompt variables

stage-sequence stage prompts can use {{task}}, {{stageName}}, {{stageLabel}}, {{diff}}, {{checksJson}}, {{previousStagesJson}}, {{previousOutputs}}, {{sharedStateJson}}, and {{contextJson}}. A stage can pass state to later stages by outputting a JSON object in a fenced afk-shared-state block. The same variable list is exported as STAGE_SEQUENCE_PROMPT_VARIABLES.

See examples/custom-pipeline-plugin for a minimal plugin-provided pipeline.