Claude Code Workflow Research — Productivity skill for Claude Code
Research notes and captured artifacts for Claude Code workflow orchestration.
How to install Claude Code Workflow Research
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open TokenRollAI/claude-code-workflow-research and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Claude Code Workflow Research does
Research notes and captured artifacts for Claude Code workflow orchestration.
Alternatives in Productivity
- Om Correct — Sweep a corrected fact through the vault — finds every note restating it, applies the correction at the single 4.6k ★
- Pi Subagents — Claude Code like Sub-Agents & Workflow Orchestration for Pi — parallel execution, live widget, fleet view, cus 967 ★
- /workflow Command — Execute complete Research → Plan → Implement workflow using chief-architect orchestration 174 ★
README
Claude Code Workflow 实现原理:从抓包看动态多代理编排
本文基于一次 Claude Code workflow 的 Reqable 抓包和本地运行产物,拆解 Claude Code dynamic workflow 的实现原理、工作机制和 agent 交接方式。
一句话概括:Claude Code workflow 的本质不是“模型在一次请求里并发思考”,而是“主模型生成一段 JavaScript 编排脚本,Claude Code 本地 runtime 执行脚本并启动多个普通子代理会话”。子代理通过 `StructuredOutput` 返回结构化 JSON,本地 runtime 再把这些 JSON 作为普通 JavaScript 值传给后续阶段,最终把脚本的 `return` 值通过 `` 回灌主会话。
证据范围
本仓库保留了同一次 workflow 运行的两类材料:
- 原始抓包: `reqable/`
- 整理后的证据: `workflow-artifacts/`
本次运行的关键 ID:
| 字段 | 值 |
|---|---|
| Session ID | 915155ec-d5cc-44ba-b5eb-fb3e513a1d8f |
| Workflow run ID | wf_1b6a0439-04e |
| Task ID | wts66al8s |
| 主题 | 桌游设计锦标赛:并行发明概念,评委团打分,综合冠军设计稿 |
关键证据文件:
| 文件 | 说明 |
|---|---|
| `workflow-artifacts/workflow-tool-prompt.md` | 主模型看到的 Workflow 工具提示词,包含触发规则、JS DSL、并发语义、恢复规则和质量模式。 |
| `workflow-artifacts/workflow-tool-schema.json` | 主模型看到的 Workflow 工具 schema。 |
| `workflow-artifacts/main-workflow-tool-use.json` | 主模型实际发出的 tool_use: Workflow,其中包含完整内联 JS。 |
| `workflow-artifacts/boardgame-tournament-wf_1b6a0439-04e.js` | Claude Code 持久化保存的本次 workflow JS。 |
| `workflow-artifacts/subagent-system-prompt.md` | 注入 workflow 子代理的系统提示词。 |
| `workflow-artifacts/structured-output-schemas/` | agent(..., { schema }) 传给 StructuredOutput 的 JSON schema。 |
| `workflow-artifacts/subagent-prompts/` | 去重后的 21 个逻辑子代理 prompt。 |
| `workflow-artifacts/examples/` | 子代理 StructuredOutput 工具调用和 tool_result 回灌样例。 |
| `workflow-artifacts/journal.jsonl` | 本地 workflow runtime 的 agent start/result 日志。 |
| [`wo |
Related Skills
Todofig
Compare a captured-figure deck against an outline document; produce a prioritized TODO of gaps (P0/P1/P2).
Pointely Search
Search Pointely for a captured team playbook matching a natural-language task.
Brain Graduate
Promote a captured idea in the second brain into a full project note with tasks.
Hatch3r Debug
Standalone debug-and-fix workflow — add strategic debug logging, collect runtime logs from the user, perform r
Quality Check
Score recent Shipwright artifacts against core rubric dimensions to detect quality drift over time.
Generate All
Generate all project planning artifacts from PROJECT-CONTEXT.md using the full orchestrated workflow
Related Agents
Research Auditor
Separate-context auditor for the kit's degraded research path. Verifies that every claim in brainstorm_notes.m
Dispatch Auditor
Process audit of the lead's own orchestration — reads the session's hook-captured dispatch ledger, checks each
Workflow Design
Read-only workflow designer for orchestration, skill boundaries, prompt structure, and durable handoff artifac