spotify-releases: notified checkpoint not persisted on partial failure → duplicate notifications

[kjgbot]

Where

spotify-releases/agent.ts (checkpoint persistence)

if (delivered && failed.length === 0) {
  await saveLastCheck(ctx, today());
  if (releases.length > 0) {
    await saveNotified(ctx, [...notified, ...releases.map(releaseKey)].slice(-500));
  }
} else {
  ctx.log?.('warn', 'spotify-releases.checkpoint-not-advanced', { since, failedArtists: failed.length, delivered });
}

Problem

Both saveLastCheck and saveNotified are gated on delivered && failed.length === 0. When messages were delivered (delivered === true) but some artist fetch failed (failed.length > 0), saveNotified is skipped — so the already-notified releases are re-sent on the next run (duplicate notifications).

Suggested fix

Persist notified whenever delivered is true (records what was successfully sent), and keep advancing lastCheck gated on delivered && failed.length === 0 so failed artists are rescanned:

if (delivered && releases.length > 0) {
  await saveNotified(ctx, [...notified, ...releases.map(releaseKey)].slice(-500));
}
if (delivered && failed.length === 0) {
  await saveLastCheck(ctx, today());
}

---

Found by CodeRabbit while reviewing vendored snapshots in AgentWorkforce/skills#87.

[kjgbot]

@kjgbot, Factory needs clarification before dispatching 101.
Reason: low-confidence triage: No repository route matched; escalate for human routing.
Question: Which repository or repositories should handle this issue? For "spotify-releases: notified checkpoint not persisted on partial failure → duplicate notifications", please reply with: (1) the exact user flow—where it starts, required inputs/actions, and the successful result; (2) permissions, validation, failure behavior, important edge cases, and anything out of scope; and (3) observable acceptance checks or tests. Say "use reasonable product defaults" for anything Factory may decide. After an authorized GitHub reply, Factory will dispatch agents; successful work will be opened as a pull request.
Authorized responder: @kjgbot (the issue reporter).
Reply with a comment that starts with [factory-reply:factory-triage-jc1zzb].