Update to MAP Audio Tools

{
  "description": "Analyse an attached recording's tempo, key, loudness against streaming targets, tonal balance, stereo and sections with MAP Audio's own engine, compare it with its source or reference tracks, play it with its waveform and spectrogram, loop and annotate its timeline, master it with Master Sculpt, process it or a built-in loop through an MFX effect preset, open PAM, MAP Audio's sampler and sequencer, to play, change and export a beat in the conversation, or render MAP Audio presets, drum hits and CM-FX and CMI programs through the Music API.",
  "included_files": [],
  "name": "map-audio",
  "skill_md_contents": "---\nname: map-audio\ndescription: Analyse an attached recording's tempo, key, loudness against streaming targets, tonal balance, stereo and sections with MAP Audio's own engine, compare it with its source or reference tracks, play it with its waveform and spectrogram, loop and annotate its timeline, master it with Master Sculpt, process it or a built-in loop through an MFX effect preset, open PAM, MAP Audio's sampler and sequencer, to play, change and export a beat in the conversation, or render MAP Audio presets, drum hits and CM-FX and CMI programs through the Music API.\n---\n\n# MAP Audio\n\n## Analysis\n\nUse `analyze_audio` when the user asks about a recording's tempo or BPM, key, loudness or LUFS, how loud it is\nfor Spotify, Apple Music or YouTube, true peak or clipping, low end or brightness, stereo width or mono\ncompatibility, or its sections. Pass the attachment or an `audioId`. Quote its numbers and notes instead of\nestimating, offer `tempo.alsoReadAs` when the user hears the tempo at half or double, and say a value is\nunavailable when it is null. It reads WAV, AIFF, FLAC and MP3 up to 32 MiB and ten minutes; for anything else,\npass on its error. Measurements do not mean you listened to the recording, and they cannot identify a song,\nan artist, lyrics or a genre.\n\n`analyze_audio` takes one extra per call. `compare` (an attachment) or `compareId` (an `audioId`) is the recording this\none was made from, such as the dry take behind a processed file; the result adds `comparison`, whose verdict says\nwhether the two are identical, differ in level only, by a linear change, are out of alignment (call again with the\n`latency` it suggests) or differ otherwise. `references` or `referenceIds` (one to three reference masters) add\n`triage`, where the mix leaves their envelopes, worst first. `spectrogram: true` adds a PNG and a series of 16 windows\nof band energy and stereo correlation; `windows` (1 to 64) gives the series alone. In ChatGPT a compared file may\nrun up to five minutes, and a mix and its references up to three each (less above 44.1 kHz).\n\nAfter an analysis, offer its `suggestions`: `open_pam` with a suggested preset at the recording's tempo, or\n`apply_mfx_preset` with a suggested MFX preset, passing its `bpm` when it has one.\n\n## Recordings\n\nUse `open_audio_workspace` to open the player. Use `open_demo_audio` when the user\nwants to try the supplied synthetic example without attaching a recording. For a recording attached by the user,\ncall `display_audio_file` with the host-provided file object. Reuse the returned\n`audioId` for later playback, region and annotation changes. Never invent a download\nURL, file ID, audio ID or server path. `read_audio_chunk` is an app-only byte-transfer\ntool; audio bytes should not be placed in conversation text. ChatGPT's desktop app calls\n`view_audio_file` itself when the user opens a WAV, FLAC, MP3 or M4A file in the file viewer;\ndo not call it.\n\nThe player analyses the recording it opens and publishes that analysis's summary, the playback\nposition, selected regions and annotations to the conversation. Use that context when available;\ncall `analyze_audio` for the full measurements, or say a measurement is unavailable instead of\ninventing values. Measurements and waveforms do not establish that you listened to a recording.\n\n## PAM\n\nUse `open_pam` when the user wants to make or hear a beat, or to try PAM. When they\nname a style, mood or tempo, call `list_pam_presets` first and pass the closest preset's\nslug; `open_pam` also takes a variation from 1 to 8 (with a preset) and a tempo in BPM. A new PAM\nstarts silent: tell the user to press Play inside PAM. The widget reports the loaded preset,\ntempo, variation and whether PAM is playing. Describe only what it reports.\n\nWhen `open_pam` returns a `pamId`, the control tools act on that PAM: `get_transport`,\n`start_transport`, `stop_transport`, `full_stop_transport`, `set_bpm`, `load_variation`,\n`cancel_queued_variation`, `randomize` and `load_preset`. Pass the `pamId` of the PAM the user\nmeans. `start_transport` works only after the user has pressed Play inside that PAM once.\n\nWith the same `pamId`:\n\n- `set_parameters` sets PAM's parameters by id. Look ids, ranges and units up with `get_parameter_info` first. A value\n  outside its range is refused, and so is the whole call, with nothing written. An effect's `cellN_<effect>Enabled`\n  flag takes 0 or 1. A `macro_*` id does not switch its macro on. The answer says which values PAM holds. A preset\n  still loading in PAM, or one that starts loading during the write, refuses the call: make it again once it has loaded.\n- `load_sample` loads an attachment or an `audioId` into a cell (`cellId` 1 to 8): up to 32 MiB and five minutes.\n- `export_audio` records what PAM plays, 1 to 16 bars (default 4), as a WAV with its own `audioId` and a one-hour\n  download link. It records in real time from the next bar line and starts a stopped transport, so it needs the user\n  to have pressed Play once. Bars that would run past five minutes at PAM's tempo, or past the 96 MiB the app stores,\n  are refused before anything records: ask for fewer. `export_preset` saves the open preset as a `.preset` file with a one-hour download link;\n  no audio tool opens it.\n\nWith the same `pamId`, `get_cmfx` and `set_cmfx` read and write a cell's CM-FX effect program, and\n`get_cmi` and `set_cmi` a cell's CMI instrument program. Call `list_cm_operators` for the cmfx/1\nformat, its operators and the factory effects and starter instruments, start from the nearest one,\nand check a new program with `check_cm` before writing it. A program the compiler refuses comes\nback with every problem and is not written.\n\nPAM in ChatGPT runs its browser engine with MAP Audio's presets. It cannot host plug-ins, run PAM's AI\nmodels or control the desktop plugin. Calling `open_pam` again opens a second PAM; when one starts playing, the\nothers pause.\n\n## MFX\n\nUse `apply_mfx_preset` when the user wants a recording processed with effects (a vocal chain, drum bus,\nreverb, distortion, a mix-bus treatment) and does not need MFX itself open. Call `list_mfx_presets` first,\nnarrowed with a `category` or `query` from the request, and pass the closest preset's slug, with the\nattachment or an `audioId`; a variation from 1 to 8 and a tempo for synced effects are optional. Recordings\nup to 32 MiB and five minutes (about three minutes of WAV). The widget renders the audio, lets the user compare it with the original and download\nit, and reports the processed copy's own `audioId`: pass that to `analyze_audio` to measure it or `display_audio_file` to show it.\nUntil the widget reports it, the processing has not finished; describe only what it reports.\n\nTo hear a preset without a recording, pass `loop` instead: `break`, `bass`, `synth`, `vocal` or `pulse`, with\n`bars` (1 to 16, default 4) or `seconds` (up to 30), never with `bpm`. The widget renders and plays the loop and\nstores nothing: there is no `audioId` or download.\n\n## Mastering\n\nUse `master_audio` when the user asks to master a mix, make it louder for streaming or match reference tracks.\nPass the attachment or an `audioId`, and up to three references (attachments or `referenceIds`). The defaults are\n-14 LUFS and a -1 dBTP ceiling; `lufs` and `ceilingDbtp` change them. It reads mono or stereo WAV, AIFF, FLAC and\nMP3 up to 32 MiB and three minutes at 44.1 kHz (less at higher sample rates), references too, and stores a 24-bit\nmaster under its own `audioId` with a download link. The player opens the master with the report under it (moves,\nloudness, each constraint as measured, and the references' findings, each with a button that asks for a master that\nholds it); `analyze_audio` measures it further. Quote the result's moves, levels and each constraint's predicted and measured value, and read its\nnotes out: a true-peak limiter holds the ceiling, checked at 16x oversampling, so a loud target is reached by limiting;\n`limiterReductionDb` says how hard it limited, and the notes name a target the renders did not reach.\n\nReferences change nothing on their own. The result's `triage` lists where the mix leaves their envelopes, worst\nfirst; to act on a finding, call `master_audio` again with the same files and a constraint of bound `envelope`,\nsuch as `{id: \"top\", metric: \"bandLevel\", channel: \"mid\", bands: [4000, 12500], bound: \"envelope\", hardness:\n\"soft\", weight: \"agreement\", reason: \"brighter than the references\"}`. Every constraint needs a `reason` that\nsays what the user asked for. At most four constraints may be hard, counting a hard loudness target; bound section classes or the whole track, not a\ntime range.\n\n## Music API\n\n`search_presets`, `get_preset`, `preset_render`, `synth_drum`, `cm_render`, `pattern_generate`, `get_credits`,\n`get_job` and `cancel_job` use the MAP Audio Music API. They work only for an account with Music API access, which is\ninvite-only. Use `preset_render` when the user wants a PAM factory preset rendered to WAV: find its slug with\n`search_presets` (web PAM's preset slugs are not the same list). It draws on the account's own render-minutes, which\n`get_credits` reports. `synth_drum` renders drum hits with PAM's drum synths. `cm_render` renders a cmfx/1 program,\na CM-FX effect over a test signal (`stimulus`) or an instrument from `notes`, and returns its level, spectral\ncentroid, tail, trips and warning codes, with no PAM open and no render-minutes: check the program with `check_cm`\nfirst, since a refused one renders nothing and the answer does not list its problems. Quote its measurements; they\ndo not mean you listened to it. `pattern_generate` returns drum and bass patterns; its MIDI file stays with the\nMusic API account. A render still running after 45 seconds returns\n`status: \"pending\"` with a `jobId`: call `get_job` with that value as `id` until it finishes, then the audio opens in the player under its\nown `audioId`. An identical call within five minutes returns the same job, so change the `seed` for another take.\nQuote the licence sentence the result gives. When a tool says the account has no access, has not accepted the terms\nor has no render-minutes left, say so, with the documentation link when the tool gives one. When it says the\nconnection was approved before it covered the Music API, tell the user to disconnect MAP Audio in ChatGPT and connect\nit again.\n\n## Scope\n\nThe player, `analyze_audio`, the PAM and MFX preset lists, `open_pam`, `get_parameter_info`, `list_cm_operators`,\n`check_cm` and `apply_mfx_preset` on a built-in loop or the example need no sign-in. Controlling or exporting from PAM, MFX on the\nuser's own recordings, `master_audio` and the Music API tools need the user's MAP Audio account. Called signed out,\nsuch a tool answers that the user is not signed in, and ChatGPT offers the sign-in: tell the user, then call it\nagain once they have signed in. Signed out, `open_pam` returns `control: \"sign_in\"` and no `pamId`; the user can still\nplay PAM by hand, and a PAM opened after signing in can be controlled. No tool needs a PAM or MFX licence.\nDo not mention prices, plans, discounts or upgrades, and do not offer to buy anything. MAP Audio does not make\nsongs, vocals or sounds from a text prompt, and has no checkout, activation or model downloads here. Do not call\ntools that the connected server does not expose.\n\nAudio IDs and download links are private capabilities and expire after one hour.\nKeep them in this conversation. Ask for a fresh attachment when a file expires.\nDirect support questions to https://map.audio/contact and privacy questions to\nhttps://audio.map.audio/privacy.\n"
}