Tools provided
Tools the agent exposes — your AI client calls them automatically when it needs them.
- open_form — Use this the moment the user wants karaoke subtitles made from an SRT or VTT file. It returns the three questions to ask: the subtitle text, the aspect ratio (9:16 or 16:9) and the preset (classic, pop or soft). It runs nothing itself.
- run_form — Internal: invoked when the user submits the form opened by open_form. Not for direct use. Same fields and same result as build_karaoke.
- run_full — Build word-by-word karaoke subtitles from SRT or VTT text in one call: the same fields and the same result as build_karaoke. A missing text, aspect or preset is asked for by name; nothing is guessed.
- build_karaoke — Turn SRT or VTT text into an ASS file where each word lights up in turn, with a preview the panel plays. Word times come from timestamps inside the file when it has them, otherwise they are estimated from line length, and the result says which. Needs text, aspect (9:16 or 16:9) and preset (classic, pop or soft); no default is used.
- convert_format — Convert subtitle text between SRT, VTT and ASS. Everything the target cannot carry (word times, styles, positions, ids, notes, comments, empty lines) is counted in lost. Converting to ass needs aspect.
- shift_and_scale — Move every time by offset_ms and/or multiply it by speed_factor or a frame-rate pair (fps_from, fps_to). Only time values are rewritten; every other line stays as it was. A shift that pushes lines before zero is refused unless drop_before_zero is true.
- parse_captions — Read subtitle text without changing it: format, line count, length, skipped blocks, repeated-line pattern, and whether a karaoke build would use word times from the file or an estimate. The panel plays the file as it is.
- check_quality — Find overlapping, out-of-order, zero-length and empty lines, and measure reading speed and line length. Limits are judged only when you give them; no broadcaster's limits are built in.
- list_capabilities — List what this agent does, its limits, what it does not check and what it never does. Use for discovery before running a tool.
- plan_inputs — Plan the inputs for a tool before running it: returns the questions, the schema and the options for each field. Defaults to build_karaoke (text, aspect, preset).
- discover_intent — Map the intent the user picked (provided.intent: karaoke, convert, retime, inspect or check) to the tool that does it and its questions. burn_in, transcribe and sync_to_audio are not available. It does not read free text.
- get_flow_status — Panel-only status probe: the step card of the last agent tool call this server process handled, or null when there was none yet.
