Run long-horizon research and coding agent workflows with DeerFlow
Use DeerFlow to orchestrate subagents, memory, sandboxes, tools, and skills for supervised research, coding, and creation tasks that take minutes to hours.
npx skills add agentskillexchange/skills --skill run-long-horizon-research-and-coding-agent-workflows-with-deerflow
DeerFlow is ByteDance’s open-source long-horizon super-agent harness for deep exploration and efficient research flows. An operator configures the application, model providers, tools, memory, sandboxes, subagents, and skills, then runs supervised research, coding, and creation tasks that require multi-step planning and execution.
What this skill actually does
Invoke this when a user needs an agent workflow to decompose a substantial research or coding objective, coordinate subagents and tools, preserve context in memory, and work inside sandboxed execution rather than asking a chat model or normal product UI to handle the task manually.
Scope boundary: this skill is for operating DeerFlow as a concrete long-horizon agent workflow harness. It is not a generic agent-framework listing, model-provider promotion, or blanket endorsement of autonomous execution; operators should define task scope, credentials, sandbox limits, and review checkpoints before running it.
Inputs and prerequisites: DeerFlow application, supported LLM provider credentials, configured tools, memory, sandboxes, and optional skills/subagents.
Setup notes: Use the upstream one-line setup or Docker quick start from the DeerFlow README, configure model and tool credentials, start with sandboxed low-risk tasks, and add review checkpoints before enabling long-running research or coding workflows.
Source and verification boundary: use https://deerflow.tech as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Multi-Framework workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.