THE AGENT SKILLS DIRECTORY
Reusable skills for AI agents. Install any of them into a Cohub world's agent with one command.
npx -y @neta-art/neta-search-cli@latest skills find "<what you need>"
Wrote a skill? Publish it here →
Basic Cohub abilities. Most agents already have these mounted; listed here for reference.
Paste this into your agent's chat — it fetches the skill and follows it.
Use the skill: auralis-thunder. Run `npx -y @neta-art/neta-search-cli@latest skills use @amarilyssage/auralis-thunder --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.npx -y @neta-art/neta-search-cli@latest skills use @amarilyssage/auralis-thunderBring a song into a browser studio that measures it, helps you master it, and gets it ready to release.
# Auralis Thunder A browser studio for any artist or band. It has no house band: the room opens neutral, an identity is optional, and every measurement comes from the file in front of you. The studio lives at `works/webs/auralis-thunder`. It is a static Work served from `index.html`, with a Node backend in `server/` for the live community room. ## The one rule Measure the file. Never invent a tempo, key, chord, vocal note, loudness or peak. Say **measured** when it came from the audio, and **proposal** when it is a suggested pass. A raw file is not a finished master; a preview is not a measurement. ## Opening the room 1. Open the Work, or the community app served by the backend. 2. Import an MP3, WAV or FLAC. The title comes from the file; the import measures tempo, key, chords, peak and the beat grid. 3. The Master rack analyses the file after import. The file lives in the browser tab — audio is not stored in localStorage, so a reload needs a re-import. 4. No audio means ask for a file. Do not fabricate a session, a waveform or a master. ## Rooms and what they do - **Home** — the session, the room profile and the first pass. The room picker sets the master's starting tune/pitch, not a band. - **Sessions** — saved session structure, lyrics, scenes and release plans. Audio is not stored. - **Auto** — a preview pass: form, chords, lyrics timing, balance, scenes, QC. Every move is a proposal until applied. - **Tracks** — the arrangement, beat/bar grid, form blocks, console balance and harmony. Import stems here. - **Master** — auto gain, auto EQ, auto tune and pitch; the manual rack; the A/B monitor; the rendered WAV. This is the room where the file becomes a record. - **Lyrics** — the sheet and film timing, auto drafts, form locking and snap-to-form. - **Visuals** — scenes, stills, grade, poster and the film. - **Identity** — an optional band or artist profile (palette, voices, DNA, rules). V.VOLT is a filled example, not the default. - **Releases** — cover, checks, deliverables and the press kit. - **Community** — auto account, twelve-slot board, live line, profiles, following, notices, reports and moderation. ## Master workflow ### Auto Auto Gain, Auto EQ, Auto Tune and Auto Pitch measure the file. The auto curve is corrective; tune and pitch only apply where a stable voice is found, and they never turn a rap or spoken take into a melody. ### Manual The manual rack is real DSP applied after the auto pass on render: - **Manual gain** — its own switch and a −12 to +12 dB fader. On, it replaces the auto makeup. - **Parametric EQ** — up to eight bands, each **Bell, Low shelf, High shelf, Notch or Tilt**. - Drag a node: x = frequency (30 Hz–18 kHz), y = gain (±12 dB). - Shift-drag or wheel = Q; the band editor also has a Q slider. - Double-click the curve or press + to add; Remove deletes the selected band (minimum one). - Shelves use real Q-driven slopes (RBJ coefficients); the notch is a true notch with depth from its gain; tilt is two opposing shelves. - The drawn curve is an approximation of the IIR response. The audio is the truth. - **Rack presets** — save the manual gain and full EQ as a named preset, recall or delete it. Presets save to the cloud when the account is linked, and follow the sync code to another device. ### A/B The A/B monitor renders both the auto chain and the manual rack from the same buffer and master chain, matches them to the quieter LUFS with the −1 dBFS ceiling, and crossfades two decks at one playhead. The **Blind test** hides the mapping until the listener picks A, B or Too close, then reveals it and keeps a running blind score. ### Stems Import one file per lane in Tracks. Each lane has its own gain, three-band EQ, tune and pitch, mute and solo. The lanes sum into one master; the global manual chain runs on the sum. Stems must share a sample rate; audio is not stored, so a reload needs a re-import. ## The release A release is one deliverable: the finished master. The chain is arrangement → balance → tuning → EQ → dynamics → effects → mix → master → QA. Target −14 LUFS integrated and true peak at −1.0 dBFS or lower. No dead air, no missing part, no clipped word. If the chain stops, say so — do not label a work in progress a single. The producer check audits the session across form, lyrics, vocal, master, release, scenes and identity. A missing band is not a miss; an instrumental or mastering room does not need a vocal or lyrics. ## Community backend - `server/index.mjs` — Node server, SQLite (`server/data/auralis.db`), content-addressed objects (`server/data/objects/`), CORS, SSE and rate limits. - Accounts are automatic: a tag and a sync code, no password. The sync code is a bearer key — never print it unnecessarily and never put it in a public reply. - The board has twelve slots; the live line shows finished records with play, like and comments. - Reports enter a moderation queue. The admin key is `server/data/admin-token.txt`; only the owner uses it. Never expose it, and never publish it inside a skill, Work or page. - **Objects** are served through the space CDN when configured; the local content-addressed cache stays for backups and fallback. - **Backups** live in `server/data/backups/`; each snapshot carries a manifest and a restore drill. `node server/backup.mjs snapshot | list | drill | restore` is the CLI. ## Honest limits - Audio is local to the tab, never in localStorage. Reload means re-import. - The manual EQ curve is an approximation; shelves, notch and tilt use real IIR/biquad audio. - Tune and pitch run on the full file for a single mix, or per lane for stems; a stem is cleaner than a full mix. - The global manual EQ applies to the master sum on stem sessions. - The community CDN is public unless R2/S3 signed storage is configured. - SQLite, the CDN and the backend are sandbox infrastructure; backups are on the same disk unless offsite storage is configured. - V.VOLT is an optional example identity, not the house band. ## Editing and verification - Change the source in `works/webs/auralis-thunder`, not only a published copy. - Check JavaScript with `node --check`. - Run the studio locally (`python3 -m http.server 8124`) and the backend on port 3000 (`PORT=3000 node server/index.mjs`). - Test at 1440, 820, 390 and 360: no sideways scroll, no console errors, the transport and the assistant intact. - Publish the Work with `node /mods/neta/neta work publish web works/webs/auralis-thunder`; verify the public URL and the cover. - If the backend changed, restart it, confirm `/api/health`, and recheck the community, media and SSE. - Never reset a room's data, rotate a key or migrate storage without the creator's approval. ## Ask when - There is no audio: ask for the file; do not invent a session. - A destructive step would touch the community (reset, hide, migrate storage, rotate the admin key): confirm first. - A storage move needs R2/S3 credentials: ask for endpoint, bucket, region, access key, secret and CDN; do not invent them. ## References - `references/release-gates.md` — the checks a finished record has to pass.