Overview

Strategies in AfferLab can export a small set of lifecycle hooks.

These hooks are invoked by the worker runtime at different stages of a conversation, allowing strategies to control prompt construction, handle tool calls, write memory, and perform cleanup.

---

onContextBuild

This is the required hook.

It runs before the host starts the model stream and must return the prompt payload for the turn.

Most strategy logic begins here.

---

onInit

If exported, onInit runs when the worker loads the strategy for runtime use.

This hook runs once per worker instance and can be used to initialize worker-local state.

It is not related to the settings page or strategy installation.

---

onTurnEnd

If exported, onTurnEnd runs after the host finishes writing the assistant message and prepares the turn-end context.

This is typically used for:

• memory writes
• background updates
• cleanup-style post-processing

---

onReplayTurn

If exported, onReplayTurn runs when the host replays existing turns during a strategy replay job.

Replay jobs may occur when a conversation switches strategies or when background reindexing is required.

---

onToolCall (coming soon)

If exported, onToolCall allows the strategy to handle a tool call inside the strategy runtime.

If this hook is not exported, AfferLab falls back to the host's default tool execution path.

---

onCleanup

If exported, onCleanup runs when the worker instance is disposed.

This hook can be used to release worker-local resources.

---

onError

If exported, onError runs when a worker-side request throws an error.