Practical tips for getting the most out of Codev and Agent Farm.
Add --dangerously-skip-permissions to your af-config.json to reduce permission prompts:
{
"shell": {
"architect": "claude --dangerously-skip-permissions",
"builder": "claude --dangerously-skip-permissions"
}
}Warning: Only use this in development environments where you trust the AI's actions.
If the Architect or a Builder crashes, you can pick up where you left off using Claude Code's /resume command:
/resume
This restores the previous conversation context so you don't lose progress.
When running 3-way reviews, launch all consultations in parallel:
consult -m gemini --protocol spir --type pr &
consult -m codex --protocol spir --type pr &
consult -m claude --protocol spir --type pr &
waitThis runs all three in the background simultaneously, saving significant time.
By default, the consult command runs in the background. If you want to watch a consultation happen in the dashboard terminal:
af consult -m gemini --protocol spir --type spec
Instead of:
consult -m gemini --protocol spir --type spec
The af consult variant runs in a visible dashboard terminal so you can observe the model's analysis.
Spawn a builder directly from an issue number:
af spawn 42 --protocol spirThe builder gets its own isolated git worktree, automatically receives the spec and plan context, and starts implementing immediately.
Use the --type flag to get focused review prompts for each stage:
consult -m gemini --protocol spir --type spec # Specification review
consult -m codex --protocol spir --type plan # Plan review
consult -m claude --type integration # PR integration reviewAvailable types: spec, plan, impl, pr, phase, integration
For large PRs, prepare context upfront and pass it to consultants:
# Prepare a summary of changes
cat > /tmp/overview.md << 'EOF'
## PR Summary
- Added user authentication
- Modified 5 files
- Key changes: JWT tokens, middleware
EOF
# Run parallel mediated reviews (~30-60s vs 2-4min)
consult -m gemini --protocol spir --type pr --context /tmp/overview.md &
consult -m codex --protocol spir --type pr --context /tmp/overview.md &
consult -m claude --protocol spir --type pr --context /tmp/overview.md &
waitCodev templates (protocols, roles, etc.) can be customized by editing files in your local codev/ directory. Local files always take precedence over the embedded skeleton.
For example, to customize the consultant role:
# Edit directly - the file already exists in your project
vim codev/roles/consultant.mdAccess Agent Farm from any device via cloud connectivity:
af tower connectRegister your tower with codevos.ai for secure remote access from any browser.
Configure porch checks for your language stack in af-config.json:
{
"porch": {
"checks": {
"build": { "command": "cargo build" },
"tests": { "command": "cargo test" },
"e2e_tests": { "skip": true }
}
}
}Works with Python (uv/pytest), Rust (cargo), Go, and any language with a build/test command.
Builders work in isolated git worktrees. Their changes don't affect your main branch until they create a PR and you merge it.
af workspace stop # Kill any orphaned processes
af workspace start # Fresh startIf things are really stuck, restart Tower:
af workspace stop # Stop Tower and all shellper sessions
af workspace start # Fresh startIf you're having port issues across multiple projects:
af ports list # See all port allocations
af ports cleanup # Remove stale entriesView the Agent Farm database state:
af db dump # Dump local project database
af db dump --global # Dump global port registry
af db stats # Show database statisticsThe Architect agent has detailed knowledge of the Agent Farm UI internals, including the Shellper session manager and Tower dashboard. You can ask it to do interesting things like:
- Debug terminal rendering issues
- Inspect shellper session state
- Create custom dashboard configurations
- Manage builder lifecycle