funasr-task-manager

Sub-agent skill for monitoring batch transcription progress and sending periodic notifications via message tool. Binds to a task_group_id, polls status at fixed intervals, and sends progress/completion/error notifications using fixed templates. Designed to run as a delegated sub-agent so the main agent can continue accepting new tasks.

Recognize transcription intent and orchestrate task submission when audio/video files or ASR keywords appear in a channel. Use when: audio/video files arrive, user mentions transcription/ASR/subtitle, user asks to batch-process media, or agent needs to proactively guide a user into the transcription pipeline.

Emergency stop workflow for FunASR Task Manager. Use when the user asks to immediately stop running transcription tasks, release occupied server slots, clear zombie TRANSCRIBING/DISPATCHED segments, diagnose active slot usage, or recover from scheduler pollution after canceled batches.

Bootstrap the funasr-task-manager development environment from scratch. Use when: first-time setup, checking if repo is cloned, installing dependencies (Python/Unicorn or Docker), or starting backend/frontend services for the first time.

Scan server-local directories for audio/video files, batch-submit them for transcription, monitor progress with proactive feedback, archive results, and handle retries. Use when: user requests batch transcription of local files, mentions scanning directories, inbox, or wants to retry failed items.

Pre-check audio/video files before transcription submission. Use when: validating whether a file can be transcribed, diagnosing file metadata or duration issues, estimating processing time before committing, or checking ffmpeg/ffprobe availability.

Reset the funasr-task-manager local database and storage to a clean state. Use when: preparing for tests (pytest, E2E, benchmark), user asks to clear tasks/data/uploads, wipe the database, rebuild after corruption, free disk space, or reset to initial state. Trigger on: "清空数据库", "重置数据", "清空任务", "只保留服务器", "其他都清空", "彻底清理", "释放空间", "reset db", "clean state", "wipe data".

Monitor transcription tasks and deliver results with quality checks. Use when: task_id or task_group_id needs monitoring or result return, user asks to export txt/json/srt/zip, user requests quality check on transcription output, or results need re-delivery for an existing task.

Safely benchmark FunASR servers and calibrate scheduling baselines. Use when: user explicitly requests benchmark or performance test, external orchestrator triggers idle-time calibration, or user asks to calibrate a newly registered server. NOT for passive use during normal transcription — missing rtf_baseline should use default RTF.

Build, run, or maintain browser-based end-to-end tests for the funasr-task-manager project. Use when Codex needs to simulate a real user dragging audio or video files into the web upload page, creating a batch of transcription tasks, waiting for scheduler execution, and validating that the system returns reasonable recognition results with saved test artifacts.