Lifecycle Hooks

Lifecycle hooks are supporting AI steps that run around normal task execution. They let OpenVibely prepare context, recall memory, select skills, and learn from completed work without turning every task into a manual setup process.

The important rule: hooks support the task. They do not replace the user's prompt, selected model, assigned agent, review workflow, or worktree boundary.

The Hook Path

route_task
-> before_run
-> normal task execution
-> after_complete

Hook	When It Runs	User Impact
`route_task`	Before the main task starts	Selects relevant standalone or agent-owned skills for the run.
`before_run`	Immediately before execution	Adds hook-provided context such as recalled project memory.
Normal execution	The primary task run	Runs the user's prompt with the selected model and, when chosen, the assigned agent.
`after_complete`	After a terminal result	Lets Memory Curator update durable project memory and Skill Curator improve relevant skills.
`scheduled`	On a schedule	Runs visible system maintenance such as memory consolidation or skill library upkeep.

What Users See

UI Area	What It Shows
Task lifecycle activity	Supporting steps that ran before or after the task.
Selected skills	Which reusable skills were prepared for the run.
Memory behavior	Relevant project memory may appear as background context for future tasks.
Scheduled system tasks	Maintenance work is visible instead of hidden.
Task result	The primary output remains the normal task execution result.

Memory Hook Behavior

Memory Curator owns memory recall, update, and consolidation. Before a task, it can recall a compact set of relevant project facts. After completed work, it can create or update durable memory when there is something worth keeping. On schedule, it can consolidate the memory library so it stays useful.

Memory is for durable project knowledge, not raw transcripts, secrets, or one-off temporary context.

Skill Hook Behavior

Skill Curator owns skill routing and learning. Before a task, it can select reusable skills that fit the prompt and execution scope. After a task, it can decide whether completed work taught OpenVibely a repeatable procedure worth saving.

Skill scope matters. No-agent tasks route standalone skills. Assigned-agent tasks route only that agent's skills. Project skills can override global skills with the same key.

Boundaries

Hooks do not give a normal task unlimited permissions.
Hooks do not turn temporary attachments into durable memory by default.
Hooks do not write arbitrary agent-owned skills outside the task's assigned-agent scope.
Hooks do not skip review, diff inspection, or merge safety.
Hooks should be treated as context and learning support, not as a second hidden user request.

Examples

Scenario	Hook Value
A project has a repeated testing convention	Skill routing can prepare the relevant testing skill before execution.
A user previously explained a repo-specific deployment rule	Memory recall can provide that rule as background context.
A completed task uncovered a durable architecture decision	Memory update can preserve it for later tasks.
Several completed tasks reveal a reusable review checklist	Skill learning can improve a scoped skill.
Memory files become duplicated or stale	Scheduled consolidation can clean them up.

Page	Why It Matters
Task Lifecycle	Shows where hooks fit in the larger status flow.
Memory	Explains recall, update, and consolidation.
Skill Curation	Explains routing, learning, and scoped skill updates.
Scheduled Task Runs	Scheduled hooks run as visible maintenance work.
Agents Overview	Agents can define reusable behavior and lifecycle participation.