Add lifecycle hooks for tasks and the overall workflow graph

[Copilot]

Implements onSuccess and onFailure lifecycle hooks at two scopes: individual tasks and the overall workflow graph.

New types (internal/types/lifecycle.go)

• Lifecycle — groups onSuccess and onFailure hooks as task name references; nil-safe accessors on both Lifecycle and Task

Per-task hooks (Task.Lifecycle)

Run after a task transitions to succeeded or failed, before any restart logic. The hook values are names of tasks already defined in the workflow file:

tasks:
build:
command: [go, build, .]lifecycle:
onSuccess: notify-slackonFailure: send-alertnotify-slack:
sh: echo build succeeded | notify-slacksend-alert:
command: [./scripts/alert.sh, build failed]

Graph-level hooks (Spec.Lifecycle)

Run when the entire workflow exits. onSuccess fires only on clean internal completion (not on Ctrl+C); onFailure fires when any task fails without restart. Graph hooks run with a signal.NotifyContext context so that a second SIGTERM/SIGINT/SIGHUP while the hook is executing will cancel it cleanly (the hook process receives SIGTERM and is killed after the grace period).

lifecycle:
onSuccess: notify-slackonFailure: send-alerttasks:
notify-slack:
sh: echo all tasks completedsend-alert:
sh: ./scripts/send-alert.sh

All hooks are best-effort — failures are logged but do not affect task or graph outcome. The named hook task is looked up and executed via the same proc.New() mechanism used for all other tasks. JSON schema updated to reflect the new types.

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

[alexec]

@copilot fix conflicts. Make sure hooks cancel
With SIGTERM etc

[Copilot]

@copilot fix conflicts. Make sure hooks cancel
With SIGTERM etc

Conflicts resolved and hook cancellation fixed.

• Merge conflicts: restored the setTerminalTitle/ringTerminalBell calls added to main, alongside our lifecycle hook calls
• Signal cancellation: graph-level hooks (which run after the main context is already cancelled) now use signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM, syscall.SIGHUP) — so if a second SIGTERM/Ctrl-C arrives while the hook is running, the hook process receives SIGTERM (via host.go's context-done handler) and is given the grace period before being killed. Per-task hooks already used the live ctx so they were fine.