Model policy

Agent and helper machinery is configuration and spawn parameters, not a fixed pipeline. Model policy chooses which model, prompt, and tool surface fits an agent or specialized pass. Veyyon is provider-agnostic: no policy hard-codes a provider.

What exists today

• Agents via the task tool (packages/coding-agent/src/task/executor.ts). /agents opens the live hub for active and persisted agent threads.
• Explicit model policies: the session model (defaultModel, /model), profile-wide agent defaults and per-agent overrides under agent, one row per feature helper under agent.helpers, plus compaction.model. An unset agent or helper runs the session’s model and effort. Edit agents and helpers under Settings → Agents → Roster. See Compaction & project memory and Models, helpers, and profiles.
• Plan / goal modes alter prompts and tool availability (/plan, /goal). The advisor watchdog (advisor.enabled and related settings, in packages/coding-agent/src/advisor/) is a background continuous-review mechanism rather than a mode you invoke; /advisor reports and configures it. See docs/handbook/src/features/advisor.md.
• Addressed inter-agent messaging via the irc tool (packages/coding-agent/src/tools/agent/irc.ts, packages/coding-agent/src/task/irc-bus.ts): send/wait/inbox/list ops over a process-global bus. send is fire-and-forget with delivery receipts; the bus wakes an idle recipient with a real turn, revives a parked one, or injects a non-interrupting aside into a busy one, the shipped analogue of wake-now-vs-defer message routing. wait (or send await:true) observes the recipient’s reply as a real turn. Controlled by isIrcEnabled: available to every agent and to a top-level session that can still spawn agents.

No fixed pipeline

Veyyon does not enforce a staged plan → implement → verify → repair handoff. It uses lighter-weight spawn, agent-policy, and irc messaging patterns instead; you compose the stages yourself.

Pair agent choice with execution-order prompts: explore → plan → edit → verify.