Invocation cookbook

Send skill calls to your agent. Run Python commands in your terminal. Profile names and file names below are synthetic examples; create the profile first and supply files your agent can actually read.

Choose your client notation

The arguments stay the same. Change the prefix for the way you installed the skill.

These calls start a conversation with the skill. Their words after the skill name are instructions interpreted by the agent, rather than strict shell flags.

Setup and new profiles

/vf-setup
/vf-new explainer-en
/vf-new divulgacion-es

Setup finds the private voice home. New asks for its brief and topics. For fiction, tell the agent explicitly that the profile is fiction; it should create it with vf.py init --type fiction. Fiction support and its limits.

Analyze existing writing

/vf-analyze explainer-en essay-one.md essay-two.md column.md
/vf-analyze explainer-en essay-one.md column.md one by one
/vf-analyze explainer-en resume

Supply texts you wrote and explain dates, coauthors, editing and shared projects when asked. “One by one” changes how decisions are presented. Resume continues from the saved record. Understand the analysis method.

Train with varied evidence

/vf-train explainer-en
/vf-train explainer-en sampling
/vf-train explainer-en openings
/vf-train explainer-en transitions
/vf-train explainer-en data
/vf-train explainer-en from essay-one.md
/vf-train explainer-en hard

Rewrite the passage, or answer OK, skip or stop. Own-text rounds use your existing paragraph; hard rounds deliberately use typical AI style and should be weighed with care. A topic name is your own instruction; focus names are documented in the training method.

Draft, edit and review

/vf-draft default-en a 120 word email asking the team to review the attached agenda by Friday. Do not invent a meeting time.
/vf-draft explainer-en a 400 word explanation of sampling error using only the notes in sampling-notes.md
/vf-edit explainer-en essay-draft.md
/vf-review explainer-en essay-draft.md
/vf-review explainer-en essay-edited.md essay-original.md

Draft creates new prose. Edit makes small changes and reports structural issues. Review gives verdicts and does not edit. Passing the original to Review lets it check meaning too. See the workflow.

Learn from real revisions

/vf-learn explainer-en draft.md final.md
/vf-learn explainer-en essay.md

The two-file form compares draft to final. The one-file form uses Git history and reports the selected versions before learning. To select a baseline, append an actual commit from that file's repository:

/vf-learn explainer-en essay.md <actual-baseline-commit>

Replace the placeholder; do not paste it literally. Edits by a coauthor, a deadline change or a one-off word limit are not personal voice evidence.

Inspect and change rules

/vf-rules explainer-en
/vf-rules explainer-en full
/vf-rules explainer-en add introduce a new technical term after an everyday example
/vf-rules explainer-en edit R01 apply this only to explanations for general readers
/vf-rules explainer-en drop R01 this is no longer a preference
/vf-rules base
/vf-rules base add in Spanish, use «» for first level quotations

R01 is illustrative. Use an ID that exists in your profile. Every add, edit or drop needs your approval of the exact change. Language-specific shared rules belong in base.md only when they really apply across your profiles. Rule design.

Derive and maintain profiles

/vf-derive divulgacion-es divulgacion-en
/vf-train divulgacion-en
/vf-status divulgacion-en
/vf-derive divulgacion-en update
/vf-migrate explainer-en
/vf-migrate base

Derive creates a provisional profile in another language, then walks through adaptation. Update proposes later source changes in a separate file. Migrate is for old layouts, not for every update. Follow the derivation guide.

Evaluate, keep holdouts and log feedback

/vf-holdout add unused-essay.md explainer-en
/vf-holdout list
/vf-evaluate explainer-en
/vf-evaluate explainer-en fidelity
/vf-status
/vf-feedback the proposed lesson was vague; show one observable action

Holdouts stay outside learning. Evaluate writes results in your private voice home. Feedback records your words privately; it does not file a public issue. Evaluation steps.

Work in Spanish

/vf-train divulgacion-es
/vf-draft divulgacion-es escribe 300 palabras para explicar el error de muestreo. Usa solo los datos de notas.md.
/vf-edit divulgacion-es borrador.md
/vf-rules divulgacion-es

The profile selects Spanish, with its language rules for accents, opening question and exclamation marks, «» quotations and calques. A Spanish name ending in -en still denotes an English profile. Choose the suffix for the output language.

Use the main skill directly

$voice-forge
Write a short explanation in explainer-en using the facts in notes.md.

The main skill can also start from ordinary writing requests when your agent discovers it. Explicit shortcuts are useful when you want a specific mode. All sixteen skill entry points are included in the reference library.