Slash commands

All autoducks commands are issued as issue comments using the /agents prefix.

Command overview

Command	Issue type	Label required	What it triggers
/agents design	Any	—	Design agent
/agents devise	Any	—	Tactical agent
/agents execute	Feature	Ready	Wave Orchestrator
/agents execute	Feature	— (no Ready)	Tactical agent (auto-escalation)
/agents execute	Non-Feature	—	Execution agent
/agents fix	Any	—	Fix agent
/agents revert	Feature	—	Revert agent
/agents close	Feature	—	Close agent

Directive arguments

All commands accept optional inline arguments:

/agents <command> [--model <model>] [--reasoning <level>]

--model

Override the Claude model for this run. Accepts any valid Claude model ID.

/agents design --model claude-opus-4-7
/agents execute --model claude-haiku-4-5-20251001

Default: value from .autoducks/autoducks.json → defaults.model

--reasoning

Override the reasoning effort level.

Value	Description
low	Faster, less thorough
medium	Balanced
high	Slower, more thorough (default)

/agents devise --reasoning medium

Pipeline shortcut

If you comment /agents execute on a Feature issue that doesn’t have the Ready label yet, autoducks will auto-escalate to the Tactical agent first, then proceed to execution. This lets you use a single command for the full pipeline without worrying about which step you’re at.

If the issue also lacks a design spec (just a raw idea), comment /agents design first to get a proper spec before planning.

Programmatic dispatch

The Wave Orchestrator and Execution agents can also be triggered programmatically — bypassing the issue comment trigger entirely. How this is done depends on the runtime.

GitHub Actions runtime (using the gh CLI):

# Trigger the wave orchestrator for a feature issue
gh workflow run autoducks-wave.yml -f feature_issue=42

# Trigger an execution agent directly
gh workflow run autoducks-execute.yml -f issue_number=43 -f base_branch=feature/42-my-feature

Other runtimes expose this through their own workflow/pipeline trigger mechanism.