Agents overview

autoducks agents are organized into adoptable layers — start with the Execution agent alone and add layers as your needs grow. You don’t need to adopt everything at once.

Layers

flowchart LR
  E["Execution
/agents execute"]
  W["+ Wave Orchestrator
parallel task dispatch"]
  T["+ Tactical
task decomposition"]
  D["+ Design
spec generation"]

  E --> W --> T --> D

  class E executionBlue
  class W waveGreen
  class T tacticalPink
  class D designPurple

  classDef executionBlue fill:#cfe8ff,stroke:#0366d6,color:#0366d6
  classDef waveGreen fill:#d4f8d4,stroke:#28a745,color:#28a745
  classDef tacticalPink fill:#f8d4e4,stroke:#e55398,color:#e55398
  classDef designPurple fill:#f0d4f8,stroke:#a836e5,color:#a836e5

Start at Execution and add layers progressively. Each layer is independent — you can use Tactical without Design, or run Execution directly without any orchestration.

Agent	Command	Type	What it does
Design	/agents design	LLM	Writes a technical spec from a feature request
Tactical	/agents devise	LLM	Decomposes the spec into numbered task issues
Wave Orchestrator	/agents execute (on Feature)	Bash	Sequences and dispatches execution agents by wave
Execution	/agents execute (on task)	LLM	Implements a single task and opens a PR

Utility commands

Command	Type	What it does
/agents fix	LLM	Retries a failed task using failure context
/agents revert	Bash	Undoes tactical planning, restores the original issue
/agents close	Bash	Full cleanup: closes issues, PRs, and branches

See the Utility commands page for full details.

LLM vs. bash

A deliberate design choice: only agents that need reasoning use LLM. Orchestration, reconciliation, and cleanup are pure deterministic bash.

Needs LLM	Pure bash
Design, Tactical, Execution, Fix	Wave Orchestrator, Revert, Close

This makes wave progression fast, cheap (no API calls), and auditable. Bash agents can be read and understood without running them.

Provider abstraction

All agents use three provider interfaces:

Interface	Prefix	Supported	Responsibility
ITS	its::	github	Issues, PRs, labels, comments
Git	git::	github	Branches, commits, merges
LLM	llm::	claude	Agent reasoning, code writing

Switching providers requires implementing the same function signatures under a new provider directory — agent logic stays unchanged. For example, adding GitLab support means creating providers/its/gitlab/ with the same function surface as providers/its/github/.

Reaction protocol

All agents use emoji reactions to report status on the trigger comment:

Emoji	Meaning
👀	Agent started (pre-execution)
👍	Agent completed successfully
😕	Agent failed

This lets you see at a glance which command triggered which run and whether it succeeded — without reading log output.