This is an early release preview. You may encounter bugs.
Hermes Dynamic Workflows logo
Unclaimed

Tool coding

Hermes Dynamic Workflows

Hermes plugin for dynamic workflows — model-written sandboxed Python orchestrating up to 1000 subagents in parallel

C 49/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Hermes Dynamic Workflows is.

Hermes Dynamic Workflows is a plugin for Hermes Agent that lets the model write a sandboxed Python script on the fly, run it in the background runtime, and orchestrate many independent subagents from it. It targets work such as codebase audits, large migrations, and cross-validated research, and it follows the dynamic-workflows pattern introduced in Claude Code.

A workflow script is async Python whose first statement is a meta literal describing the run and its phases. Inside the script, agent(prompt, opts) spawns a child agent (with options for a structured-output schema, model, agentType, and isolation="worktree"), pipeline chains stages per item without a barrier, parallel runs with a barrier, and phase, log, workflow(), args, and budget handle progress, nesting, inputs, and token budgets. The plugin persists the script and each child's full transcript and injects a into the conversation when the run completes.

Configuration in ~/.hermes/config.yaml (or HERMESDYNAMICWORKFLOWS_* variables) sets concurrency, a hard cap of 1000 agents per run, wall-clock and per-child timeouts, toolsets children may or may not use, launch-approval requirements, and completion notifications. An optional hermes-workflows terminal dashboard shows run lists, per-phase progress, and each subagent's prompt and output live.

Features

  • Model-written scripts: the agent authors a sandboxed Python workflow at request time
  • Concurrency primitives: agent(), pipeline() (no barrier), and parallel() (with barrier)
  • Agent types: built-in general-purpose, explore, plan, and verification types, plus custom types defined as Markdown files at project or user level
  • Structured output: a schema option enforces the shape of a child agent's result
  • Worktree isolation: child agents can run in their own git worktree, cleaned up by default
  • Governance: launch approval, child approval policies, blocked toolsets, and runaway guards on agent count and time
  • Persistence and history: scripts and transcripts stored in state.db, browsable with /workflows
  • Live dashboard: a separately installed hermes-workflows console for watching runs

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Hermes Dynamic Workflows.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Hermes Dynamic Workflows →

Browse the catalogue