feat: add sonilo-cli package and context7.json

[spencer-zqian]

Summary

Ports sonilo-js PR #12 to Python: a new sonilo-cli package plus a repo-root context7.json. Rebased on sonilo 0.5.0, so it also covers the new combined-soundtrack endpoints.

What is sonilo-cli for?

A command-line wrapper around the sonilo SDK, so you can generate music and SFX from the terminal without writing Python. Instead of a script:

from sonilo import Sonilo
Sonilo(api_key="sk-...").text_to_music.generate(prompt="warm lo-fi piano", duration=30).save("song.m4a")

you run:

sonilo text-to-music --prompt "warm lo-fi piano" --duration 30

Useful for quick demos, shell/CI/cron automation, and non-programmers. It's a thin layer — all capability comes from the SDK; the CLI only translates flags into SDK calls (hence zero third-party dependencies, stdlib argparse only).

What is context7.json for?

Context7 is a service that feeds up-to-date library docs to AI coding assistants so they stop hallucinating stale/wrong APIs. context7.json is the library-author config we hand Context7: which folders to skip when indexing, plus a set of agent-facing rules (read the key from SONILO_API_KEY; text-to-sfx/video-to-sfx are always async; result URLs are short-lived presigned links — don't send the auth header; which error types to catch). Once the repo is registered on Context7, anyone using an AI assistant to write Sonilo code gets our own accurate guidance instead of a guess.

What's included

• New sonilo-cli package (bin: sonilo) wrapping the SDK: account, usage, text-to-music, video-to-music, text-to-sfx, video-to-sfx, video-to-sound, video-to-video-sound, tasks get/tasks wait. Auth via SONILO_API_KEY or --api-key. Async-only options (--format wav, --isolate-vocals, --preserve-speech) submit + poll automatically instead of streaming. Zero third-party deps (stdlib argparse).
• Combined-soundtrack commands for the 0.5.0 endpoints: video-to-sound (mixed audio, default output.wav) and video-to-video-sound (source video with audio muxed in, default output.mp4). Both take --music-prompt/--sfx-prompt, --preserve-speech, --no-ducking, and a repeatable --stem music|music_processed|sfx that saves the individual layers next to the combined output (--output soundtrack.wav → soundtrack.music.m4a).
• context7.json at the repo root — folder exclusions plus Python-flavored agent rules, including the new sound endpoints, the ducking tri-state gotcha, stem semantics, and the free trial.
• CI + publishing: ci.yml installs and tests all three packages; new publish-cli.yml mirrors the video-kit OIDC trusted-publishing workflow (tag prefix sonilo-cli-v*).

Note on ducking

ducking is default-ON server-side, and build_v2s_parts only emits a boolean when it isn't None. The CLI therefore forwards None when --no-ducking is absent and False only when it's passed — forwarding a plain store_true default of False would have silently disabled the server default. Same tri-state for preserve_speech. Both directions are covered by tests that assert the field is absent from the form body when the flag is unset, not merely that it's false.

Dependency pin

sonilo-cli pins sonilo>=0.5,<0.6. The original >=0.4,<0.5 excluded 0.5.0 and would have reproduced the ResolutionImpossible that sonilo-video-kit just hit and fixed by widening to <0.6.

Not included — needs the account owner

• First PyPI publish of sonilo-cli: for a brand-new project name, a PyPI Trusted Publisher must be configured for sonilo-cli before the first tag push, or the publish step fails. That's a PyPI org-settings action.
• Registering the repo on Context7: context7.json is ready, but claiming the library at context7.com requires signing in there.

Test plan

• pytest sonilo-cli → 28 passed
• Full repo pytest (SDK) → 167 passed; pytest sonilo-video-kit → 52 passed (no regression)
• 3-package editable install resolves cleanly against sonilo 0.5.0
• Error paths: missing API key, invalid key (401 surfaced cleanly, no traceback), missing required flag, unknown command / unknown tasks subcommand, --video + --video-url together — all exit 1 with a one-line sonilo: message
• Verified live against the production API: account and usage --days 7 return real data; text-to-music --duration 5 → valid 5.06s MP4/M4A (streaming path); text-to-sfx --duration 2 → valid 2.04s 16-bit/44.1kHz WAV (async submit → poll → presigned download). Billing confirmed to apply the documented minimums (5s music + 2s sfx billed as 10s + 3s).
• video-to-sound / video-to-video-sound not live-tested (respx-stubbed only) — same submit/poll/download code paths as the tested commands

[sapient-app]

Sapient docs evals

Waiting for the staging docs URL before running evals.

Sapient will start the selected PR evals automatically when GitHub reports a successful docs deployment for this PR. This usually happens within 15 minutes.

Commit: eae756b
Status: waiting for staging docs URL