From8 specific workflowsselect the current task. Detailed steps and parameters are inrecipes.json, and actual method observations are incommand evidence，GitHub sourcerecords workflow facts and licensing; the body text and examples are original adaptations and do not copy external scripts.
Same session and standalone CLI
When the embedded editor is already open, use that MCP session, wait for editor_status to be ready, read the host schema, then call execute {"command":"方法ID","params":{…}}; do not run the CLI on another file and then claim the screen is in sync. For each step, choose the execution based on the current input and the returned object, save the source project and exported file, and reopen them to check.
CLI：soundcraft-cli run --cmd <方法ID=JSON> … --save output.scraft, each id=JSON is one argv parameter. First use soundcraft-cli info input.wav to measure the original file. Each example creates a new 48kHz empty session by default; when executing in an existing project, skip the newDocumentOnly initialization, inspect first, and select by actual track/clip ID; you must not import the example Voice and overwrite the real selection. After saving, verify with a new process run --in output.scraft --cmd session.inspect={} , and re-bounce to check the effect.session.audio_healthOnly checks references; does not analyze noise floor, clipping, or loudness.
The target parameter for AudioSuite normalize istarget_db; invert polarity isinvert, and time reversal isreverse; the two are not the same operation. Numeric position defaults to samples; use{seconds:…}to explicitly express seconds.
Execution boundaries
Each example is an independent task. Initialization, new imports, and example sets cannot be used unconditionally on existing projects. Output paths use new names, and the original file remains unchanged. Real content must be auditioned in full to confirm there is no distortion, swallowed words, or unintended silence. Structured assertions apply only to embedded synthetic input and do not replace acceptance of the effect on real material.
Skill store scenario experience
When encountering situations where cuts are processed first and then global volume, podcast chapters correspond to real audio, or independent vocals and background material must be preserved, readstore workflow adaptation, and use the corresponding Craft method and actual result checks. Upstream skills are not a runtime dependency.
Sync silence and limited loops
When preserving the sync relationship between dialogue and video, or turning low-level segments into gaps, hand off to the soundcraft-voice-silence-gaps skill. Install separately:npx skills add full-aigc-skills/soundcraft-skills --skill soundcraft-voice-silence-gaps。
When cue tones or sound effects with verified seams require an exact total count and subsequent expanded editing, hand off to the soundcraft-finite-audio-loop skill. Install separately:npx skills add full-aigc-skills/soundcraft-skills --skill soundcraft-finite-audio-loop. This batch is provided as a pre-release alongside companion plugin development; auditioning on real material and host sync have not yet been accepted.