Improve LastFM mood analysis robustness and efficiency

[err53]

Motivation

• Make the LastFM mood feature resilient to partial metadata failures so one failing track-info lookup doesn't break the entire mood result.
• Reduce LLM payload size and noise to lower latency/cost and improve mood signal quality.
• Reduce frontend/network load from overly aggressive polling for the latest track.

Description

• Replace Promise.all with Promise.allSettled when fetching per-track details and fall back to recent-track fields for any failed lookups to ensure a complete, best-effort payload.
• Build a compact per-track payload (name, artist, album, tags, trimmed wikiSummary) and pass that to the LLM instead of full Last.fm objects to reduce tokens and noise.
• Update the system/user prompt schema to document and reference the new CompactTrack shape and simplified cue names.
• Add cache options (revalidate and tags) to the unstable_cache call and change the call-site to analyzeRecentMood(input.user, input.user); reduce client polling interval in LastFM component from 1*1000 to 5*1000.

Testing

• Ran pnpm -s lint, which could not complete in this environment due to missing required environment variables (LASTFM_API_KEY, OPENROUTER_API_KEY, NEXT_PUBLIC_POSTHOG_KEY), so automated lint checks did not finish successfully.

---

Codex Task

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
personal-site	Ready	Preview, Comment	May 27, 2026 11:07pm