quranic-universal-audio

Headless one-off Inspector admin operations — wraps the same services/admin/* and state.transition the UI calls. Use for quick patches when the Reviews/Users/Permissions/Requests UI surface isn't enough — publish a finished TS job, force-unlock a released reciter, cancel + relaunch a stuck HF job, and similar one-off tasks.

Use when localizing the Inspector frontend — extracting UI strings into Paraglide messages, adding/wiring `m.*()` calls in Svelte components, authoring en/ar message JSON, translating chrome to Arabic (MSA), or verifying i18n coverage. Trigger phrases include "internationalize", "localize", "translate to Arabic", "extract strings", "i18n", "add a message key", "wire up the locale switcher".

Screenshot the real Timestamps analysis view (UnifiedDisplay) for any reciter:surah:verse straight from shards — fast, drift-free, no SPA/audio/OAuth.

Run the Inspector app locally — any mode (dev bucket / offline fixtures / live dev or prod Space), any worktree, conflict-free.

Deploy any branch to its HF Space directly from the CLI (local, no push) and monitor until the Space is live.

Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface.

Inspector audio subsystem — everything between bytes-on-disk and an `<audio>` element rendering audio in the browser, plus the extraction → bucket → inspector handoff that produces `reciters/<slug>/` in the first place. Covers both debugging AND building audio features.

Reference for running workloads on Hugging Face Jobs.

Inspector performance across the whole stack — single-worker backend (SQLite / caching / bucket-I/O / hot paths), frontend (bundle / render / wire-shape), and the live-stack runtime probing that produces the evidence (Playwright/Chrome MCP + Flask/Vite log correlation). The cost model, the measurement playbook, and the cautionary patterns.