Autopilot Mode

Autopilot mode runs an agent in a loop — implement, validate, retry — until all checks pass or the iteration budget runs out. A fresh agent session is spawned each iteration to avoid context rot.

When to use

The scope is well-defined with clear acceptance criteria
The feature spec has a ## Validation section with executable checks
You want hands-off execution

Modes

Mode	Agents	Command
Autopilot	Single agent	`feature-do 07 --autonomous`
Swarm	Multiple agents	`feature-start 07 cc gg --autonomous --auto-submit`

Swarm is Fleet + Autopilot — multiple agents compete, each running autonomously.

How it works


Iteration 1:  spawn agent → implement → run validation
                                          ↓ fail
Iteration 2:  spawn fresh agent → read previous attempt → fix → validate
                                          ↓ fail
Iteration 3:  spawn fresh agent → read attempts 1-2 → fix → validate
                                          ↓ pass ✓
Done — agent signals submitted

Each iteration gets a fresh context window but can see the history of previous attempts.

Workflow

Single agent (Autopilot)


# Create and set up the feature first
/aigon:feature-create fix-pagination
# → Describe the bug: pagination shows wrong page count after filtering...
/aigon:feature-prioritise fix-pagination
/aigon:feature-start 42
 
# Run autonomously (max 5 iterations by default)
aigon feature-do 42 --autonomous
 
# Or with options
aigon feature-do 42 --autonomous --max-iterations=3 --agent=cx

Multiple agents (Swarm)


# Set up Fleet with autonomous mode
aigon feature-start 42 cc gg cx --autonomous --auto-submit
 
# Or use the autopilot command for full orchestration
aigon feature-autopilot 42 cc gg cx

The feature-autopilot command handles setup, spawning, monitoring, and evaluation in one step.

Writing validation checks

Add a ## Validation section to your feature spec:


## Validation
 
\`\`\`bash
cd site && npm run build && echo "Build OK"
npm test
\`\`\`

Each command must exit 0 for validation to pass. The agent runs these after each implementation attempt.

Flags

Flag	Default	Description
`--autonomous`	off	Enable autonomous retry loop
`--max-iterations=N`	5	Maximum retry attempts
`--auto-submit`	off	Auto-signal submitted on success (for Swarm)
`--agent=<id>`	cc	Which agent to use
`--dry-run`	off	Show what would run without executing