Wire pause resume end to end: triggers now wake paused missions

[mcclowes]

Fixes #197

Problem

pause resumeOn: webhook registered the endpoint, the caller POSTed to it, the server recorded the event — and nothing continued the run. PauseManager.handleWebhook and startMonitoring had zero production callers, so timeout pauses likewise had no running component watching the deadline. The resume mechanism (execute({ resumeFrom }) replaying the durable log past an already-resumed pause) was sound; the trigger that invokes it was missing.

What this adds

PauseOrchestrator (src/pause/orchestrator.ts) — the wiring between the webhook server, the pause store, and a host resume callback:

• routes an inbound delivery to the waiting pause holding that path's trigger (registration id first, path as the restart-safe fallback)
• polls for expired timeout pauses, with an immediate sweep at startup so pauses that expired while nothing was running resume without waiting a tick
• recovers across restarts: re-registers webhook paths whose registrations were lost, and picks up deliveries that landed before it started
• hands each resumed pause to resume(pause); a throwing callback is reported, never an unhandled rejection, and the pause stays recoverable via a manual resume

executeWithResume(source, config) (src/index.ts) — run a mission to completion: execute, wait for a trigger, re-execute with resumeFrom, loop. Resume runs are serialized through a queue so a webhook landing while the paused run is still unwinding can't start a concurrent run. Requires config.executionLog (replaying past a pause is log-based) and throws a clear error without it.

Supporting changes: PauseManager.handleWebhookEvent (event → pause mapping), WebhookServer.onEvent() post-construction subscription, registration/event read accessors, path on delivered events, and getPort() reporting the actual bound port so port: 0 works in tests.

The test that would have caught this

src/interpreter/executor-pause-resume-e2e.test.ts: pause a mission, deliver a real webhook over HTTP, assert the pipeline tail runs exactly once — one pause.created, one pause.resumed, one mission.completed, one stored record carrying the webhook payload — and a duplicate delivery 404s without re-running anything. Plus a timeout-driven resume with no webhook at all, and orchestrator unit tests covering live routing, early delivery, restart recovery, failing resume callbacks, and the webhook-vs-timeout race resuming exactly once.

Docs updated (docs/durability/pause.md, README, CLAUDE.md): the "you drive the resume" warning is replaced with the executeWithResume / PauseOrchestrator story; plain CLI runs still exit on pause and resume via --resume.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
reqon	Ready	Preview	Jul 29, 2026 9:14pm