feature-start

Synopsis


aigon feature-start <ID> [agents...]

Shortcuts

/afs (slash command) · aigon afs (CLI)

Description

feature-start moves a feature from backlog to in-progress by creating the implementation workspace and opening agent terminals. The number of agents you pass determines the mode:

No agent — Drive mode (branch): creates a feature branch in the current repo, switches to it, and leaves you on that branch to implement manually
One agent — Drive worktree: creates an isolated worktree for that agent, opens the agent terminal pre-loaded with feature-do
Two or more agents — Fleet mode: creates one worktree per agent, opens all agent terminals simultaneously, agents implement in parallel

Always use this CLI command — never manually create branches or worktrees. The CLI records the backlog-to-in-progress workflow transition and prepares the branch or worktree; in the stable layout, lifecycle folders are generated views rather than spec-file moves.

What it does

Records the feature’s transition from backlog to in-progress and refreshes the generated lifecycle view
Creates a feature branch (feature-<ID>-<slug>) or worktree(s) (feature-<ID>-<agent>-<slug>)
For worktrees: writes a .env.local with an isolated PORT value for the dev server
Opens agent terminals pre-loaded with feature-do <ID>

Model and effort overrides

Per-run overrides (after any agents list):


aigon feature-start 42 cc --models=cc=claude-opus-4-7 --efforts=cc=high

If you omit --models / --efforts, the CLI uses the spec complexity → per-agent complexity ladder → aigon config models resolution. See Agents: Spec complexity and model defaults.

Examples


# Drive mode (branch) — manual implementation in current repo
aigon feature-start 42
 
# Drive worktree — isolated development, agent opens automatically
aigon feature-start 42 cc
 
# Fleet mode — cc and ag implement in parallel
aigon feature-start 42 cc ag
 
# Fleet mode with model overrides
aigon feature-start 42 cc ag --models=cc=claude-opus-4-7,ag=gemini-3.5-flash-medium

What to do next

Mode	After feature-start
Drive (branch)	Implement manually, then `feature-close <ID>`
Drive worktree	Wait for the agent to signal `implementation-complete`, then `feature-close <ID>`
Fleet	Wait for all agents to signal `implementation-complete`, then `feature-eval <ID>`

feature-prioritise — move the already allocated ID to backlog (run before start)
feature-do — the command agents run inside the workspace
feature-open — re-open a crashed agent terminal after start
feature-eval — compare Fleet implementations after agents signal completion
feature-close — merge the winning implementation