| name | ghost-voice |
| description | Manage the voice profile across the three GhostAI config tiers (in-repo,
project-local, global). View what's loaded, create or update at any tier,
promote a profile from one tier to another, or delete one. Use when the
author asks to "update voice", "change voice profile", "share voice with
co-authors", "promote voice to repo", or "view voice profiles".
|
| allowed-tools | ["Read","Glob","Grep","Bash","Write","Edit","AskUserQuestion"] |
/ghost-voice — Voice Profile Manager
You manage the author's voice profile across GhostAI's three config tiers.
This skill is the canonical entry point for any voice-related operation that
isn't part of initial book onboarding.
See shared/config-hierarchy.md for the tier model. The short version:
- In-repo (
{repo}/.ghostai/voice-profile.json) — committable, shareable
- Project-local (
~/.ghostai/projects/{slug}/voice-profile.json) — per-book on this machine
- Global (
~/.ghostai/voice-profile.json) — machine-wide default
Highest existing tier wins on load. No merging.
Step 1: Run the Shared Preamble
Read shared/preamble-core.md and follow its instructions. The preamble
exports $GHOST_VOICE_FILE, $GHOST_VOICE_TIER, $GHOST_TIER_REPO,
$GHOST_TIER_PROJECT, $GHOST_TIER_GLOBAL, and the helpers
ghost_tier_dir and ghost_source_label.
Step 2: Survey All Three Tiers
Don't just trust the resolver — show the author what exists at every tier
so they have a complete picture before deciding what to do:
echo "=== Voice profiles across tiers ==="
for tier in repo project global; do
dir=$(ghost_tier_dir "$tier")
file="$dir/voice-profile.json"
if [ -f "$file" ]; then
bytes=$(wc -c < "$file" | tr -d ' ')
mtime=$(stat -f '%Sm' -t '%Y-%m-%d' "$file" 2>/dev/null || stat -c '%y' "$file" 2>/dev/null | cut -d' ' -f1)
label=$(ghost_source_label "$tier")
marker=""
[ "$tier" = "$GHOST_VOICE_TIER" ] && marker=" ← active"
echo "$tier: $bytes bytes, updated $mtime ($label)$marker"
else
echo "$tier: (none)"
fi
done
For each existing profile, also read it and produce a one-line summary
(person, formality, average sentence length) so the author can compare
without opening files.
Step 3: Pick the Action
Use AskUserQuestion to offer the operations:
"What would you like to do with your voice profile?"
- A) View — show the active profile in detail
- B) Create or update at a specific tier — generate a profile from a
writing sample or analyze chapters, then save to the tier you choose
- C) Promote / copy to another tier — copy an existing profile from
one tier to another (e.g., promote project-local to in-repo so
co-authors share it)
- D) Delete at a specific tier — remove a profile, exposing whichever
tier sits below it
A: View
Print the active profile ($GHOST_VOICE_FILE) with field-level explanations.
Lead with a natural-language summary, e.g.:
"Your active voice profile comes from this book's repository
(.ghostai/voice-profile.json). It says: second-person, conversational
formality, average sentence length 18 words, occasional rhetorical
questions, frequent code-first paragraphs."
If profiles exist at lower tiers and differ from the active one, mention
them: "You also have a different profile at the global tier — it's hidden
by the repo profile. Want to view, promote, or delete it?"
B: Create or update
Ask the author for source material:
"How should I build the voice profile?"
- A) From a writing sample — paste text or share a file/URL
- B) From this manuscript's chapters — analyze 2-3 chapters
- C) From a description — tell me how you want to sound
(more casual, more authoritative, etc.)
Build the profile (extract sentence length, vocabulary, person, formality,
humor/anecdote/question patterns, paragraph length).
Then ask which tier to save to:
"Where should I save the new voice profile?"
- A) In this book's repository (
.ghostai/voice-profile.json) — shared via git
- B) Project-local on this machine (
~/.ghostai/projects/{slug}/voice-profile.json)
- C) Global default for this machine (
~/.ghostai/voice-profile.json) — used everywhere
If GHOSTAI_DEFAULT_TIER is set, skip the prompt.
If a profile already exists at the chosen tier with different content, confirm
overwrite via AskUserQuestion before writing.
Write the file:
target_dir=$(ghost_tier_dir "$chosen_tier")
mkdir -p "$target_dir"
mkdir -p "$GHOST_TIER_PROJECT"
echo "$chosen_tier" > "$GHOST_TIER_PROJECT/.tier"
If chosen_tier=repo and $GHOST_TIER_REPO/.gitignore doesn't exist, create
the recommended one (see shared/config-hierarchy.md).
After writing, run the shadow check (Step 4).
C: Promote / copy
Ask source and destination tiers:
"Copy from which tier?" — list only tiers that have a profile.
"Copy to which tier?" — list the other two tiers.
Read the source file, write byte-identical content to the destination
directory (mkdir -p first). Then run the shadow check (Step 4).
If the user says "promote" specifically (vs "copy"), offer to delete the
source after the copy completes — but ask first.
D: Delete
Ask which tier to delete from. Warn if deleting will change which profile
is active (i.e., if deleting the highest tier reveals a lower-tier profile,
or removes the only profile entirely).
Use AskUserQuestion to confirm:
"Delete the voice profile at the {tier} tier? This will {describe
consequence — 'leave you with no voice profile' or 'fall back to the
{next-tier} profile'}."
Then rm the file. If the deletion empties $GHOST_TIER_PROJECT of
everything except .tier, leave .tier alone (the anchor still applies
to learnings and reviews).
Step 4: Shadow Check (after any write or copy)
If you just wrote a profile to a higher tier and a different-content profile
exists at a lower tier, that lower one is now invisible. Use AskUserQuestion
to surface this once:
"You now have a voice profile at the {higher-tier} tier. Your
{lower-tier} profile is different and is now shadowed (it won't be
loaded). What should I do with it?"
- A) Delete the shadowed profile — the new one is canonical
- B) Keep both — I might want to switch back later
- C) Show me the diff first
If the two profiles are byte-identical, silently delete the lower-tier copy
without asking (it's redundant).
No shadow check needed for learnings.jsonl or reviews/ — those follow
$GHOST_ANCHOR_TIER rather than precedence.
Step 5: Confirm and Complete
Show the final state:
┌─────────────────────────────────────────────────┐
│ GhostAI /ghost-voice │
│ Voice profile: {action taken} │
│ Active source: {tier} · `{path}` │
└─────────────────────────────────────────────────┘
If the author created or moved a profile, suggest the natural next step:
"Run /ghost-edit on a chapter and the new voice will be applied."
STATUS: DONE
Skill: /ghost-voice | Action: {view|create|update|promote|delete} | Tier: {tier}
