feat(jobs): persist lastResult and lastRanAt per job in state

[Nibbler1250]

Resolves #175.

Adds two optional fields to each job entry in state.json so the statusline / heartbeat prompt / external monitors can see what each job last did without re-executing it.

Changes

src/statusline.ts — extend StateData.jobs[] type:

jobs: {
  name: string;
  nextAt: number;
  /** Outcome of the most recent run. Absent until the job runs at least once. */
  lastResult?: "ok" | "error" | "skipped";
  /** Unix timestamp (ms) of the most recent completion. Absent until first run. */
  lastRanAt?: number;
}[];

src/commands/start.ts — three small additions:

1. Declare an in-memory jobLastResult Map alongside the existing jobRetryState Map (same lifecycle: resets on daemon restart, no disk persistence).
2. In the runJob() .then() handler, right after the run completes (next to where jobRetryState is updated), record the outcome:

   jobLastResult.set(job.name, {
     result: r.exitCode === 0 ? "ok" : "error",
     ranAt: Date.now(),
   });

3. In updateState(), spread the recorded fields into each job entry so they land in state.json on the next 60-second tick:

   jobs: currentJobs.map((job) => {
     const last = jobLastResult.get(job.name);
     return {
       name: job.name,
       nextAt: nextCronMatch(...).getTime(),
       ...(last ? { lastResult: last.result, lastRanAt: last.ranAt } : {}),
     };
   })

Properties

• Additive: fields are absent until the job runs at least once; existing consumers see no change.
• No new disk I/O: piggybacks on the existing updateState() 60s tick.
• No new dependency: just a Map<string, { result, ranAt }> in memory.
• Lifecycle matches jobRetryState: in-memory only, drops on daemon restart (no stale debt across restarts, consistent with how retry state already works in this file).

Notes on "skipped"

I included "skipped" in the union type for forward compatibility (e.g. a future heartbeat could mark a job skipped due to rate-limit / disabled / pause), but this PR doesn't write skipped anywhere yet. Happy to drop that variant if you'd rather keep the surface tight to "ok" | "error" for now.

Build

• bun build src/index.ts --target=bun → 0.42 MB, clean.
• bunx tsc --noEmit → no new errors. The two pre-existing errors at start.ts:442 / :447 (forwardToTelegram / forwardToDiscord arity mismatch) live on master already; I left them alone.
• No new lockfile / package changes.

Source

Originally requested by myself in #14 (Mega-Post) — thanks @TerrysPOV for breaking it out into its own ticket.

[TerrysPOV]

Found 2 issues:

1. Daemon crashes at startup when any jobs are configured — updateState() is called immediately at line 770, but const jobLastResult (which the updated updateState() body now reads via jobLastResult.get(job.name)) is declared at line ~779. const variables are in the temporal dead zone until their declaration is reached; accessing one before that throws ReferenceError: Cannot access 'jobLastResult' before initialization. Any user with jobs configured will see the daemon die on startup. Fix: move the jobLastResult declaration above the updateState() call (alongside or before jobRetryState).

claudeclaw/src/commands/start.ts

Lines 750 to 757 in a7dd0d0

	heartbeat: currentSettings.heartbeat.enabled
	? { nextAt: nextHeartbeatAt }
	: undefined,
	jobs: currentJobs.map((job) => {
	const last = jobLastResult.get(job.name);
	return {
	name: job.name,
	nextAt: nextCronMatch(job.schedule, now, currentSettings.timezoneOffsetMinutes).getTime(),

claudeclaw/src/commands/start.ts

Lines 776 to 780 in a7dd0d0

	// In-memory retry state: resets on daemon restart (no stale debt across restarts).
	const jobRetryState = new Map<string, { failCount: number; retryAt: number }>();
	// Track each job's most recent outcome so state.json can expose lastResult/lastRanAt
	// for crash-recovery + status displays. Resets on daemon restart (in-memory only).

2. "skipped" is declared in the type but never written — The type union in statusline.ts includes "skipped", and jobLastResult is typed to accept it, but the only write path sets "ok" or "error" from r.exitCode. The natural trigger for "skipped" — jobs bypassed because isRateLimited() returns true — has no jobLastResult.set() call. Any consumer checking lastResult === "skipped" to detect rate-limit suspensions will never see it. Fix: set { result: "skipped", ranAt: Date.now() } inside the isRateLimited() branch of the setInterval, once per job that was due to fire.

claudeclaw/src/statusline.ts

Lines 8 to 12 in a7dd0d0

	jobs: {
	name: string;
	nextAt: number;
	/** Outcome of the most recent run. Absent until the job runs at least once. */
	lastResult?: "ok" | "error" | "skipped";

Please effect these changes, commit and resubmit.