vivadicta search and rewrite

Use this skill to locate an existing transcription in the user's VivaDicta history and apply an AI preset to it. Covers natural-language IDs, preset resolution, and the end-to-end flow.

Preconditions

• VivaDicta.app installed at /Applications/VivaDicta.app and running (rewrite requires the UI).
• User has an AI provider configured with an API key (keychain) or a local MLX model loaded.
• Some history exists. If vivadicta recent 1 is empty, there's nothing to rewrite.

Identifier forms

rewrite (and get) accept three identifier shapes. Prefer in this order:

1. Keyword: when the user clearly means the newest entry.

   vivadicta rewrite latest --preset summary
   # also: last, most recent, newest, current

1. Natural-language query: when the user describes the transcription.

   vivadicta rewrite "yesterday's call with sam" --preset action_points
   vivadicta rewrite "coding session this morning" --preset professional

1. Raw UUID: when you already have an ID from a prior command.

   vivadicta rewrite 1701DB3C-AF1F-4191-A791-5DABA81D8096 --preset email

Session handles (t1, t2) from MCP don't carry across one-shot CLI invocations - use the UUID instead.

Resolver behavior

The natural-language matcher:

• Splits the query into tokens (≥3 chars), lowercases, and searches text + enhancedText + powerModeName.
• Also extracts a timeframe hint ("yesterday", "last week") and filters by it.
• Ranks candidates; if the top score is ≥3x the second score, picks the top. Otherwise returns transcription_ambiguous (exit 4) with the top 5 previews on stderr.

When ambiguous, the error lists candidates. Re-run with a more specific phrase or with a raw UUID from the candidates list.

Preset resolution

Preset aliases are fuzzy-matched too: summary / Summary / sumary all resolve to the Summary preset. rewrite returns preset_ambiguous (exit 3) with candidates if the match is unclear.

Discover available aliases:

vivadicta presets --output json | jq -r '.presets[] | "\(.alias)\t\(.name)"'
# regular           Regular
# summary           Summary
# action_points     Action Points
# email             Email
# ...

Built-in aliases include: regular, summary, action_points, takeaways, mind_map, key_points, professional, casual, email, chat, coding, rewrite, short, elaborated, simplify, translate_en, translate_ru, translate_es, translate_zh, plus more translate_* variants and user-created presets (typically prefixed custom_).

End-to-end flow

# Discover the transcription
vivadicta search "sprint planning" --since "last week" --limit 5

# If exactly one matches, rewrite directly with a natural-language query
vivadicta rewrite "sprint planning last week" --preset action_points

# If multiple, grab the UUID from the search output
UUID=$(vivadicta search "sprint planning" --limit 1 --output json | jq -r '.transcriptions[0].ref')
vivadicta rewrite "$UUID" --preset action_points

The rewrite result prints the original + rewritten text. The rewritten version is also saved as a TranscriptionVariation on the record - follow-up vivadicta get <uuid> --all-variations will include it.

Dry-run

Preview the resolved transcription + preset without hitting the UI app:

vivadicta rewrite "yesterday's call" --preset summary --dry-run
# Action       : rewrite (dry-run)
# Transcription: 1701DB3C-AF1F-4191-A791-5DABA81D8096
# Title        : Call with Sam about the sprint plan
# Preset       : summary

Useful for confirming the resolver picked the right transcription + preset before the AI call.

Common patterns

"Make action items from today's call":

vivadicta rewrite latest --preset action_points

"Rewrite my dictation as a professional email":

vivadicta rewrite latest --preset email --output markdown

"Translate yesterday's note to Russian":

vivadicta rewrite "yesterday's note" --preset translate_ru

Chain with export (see vivadicta-vault-export):

UUID=$(vivadicta recent 1 --output json | jq -r '.transcriptions[0].ref')
vivadicta rewrite "$UUID" --preset summary > /dev/null
vivadicta get "$UUID" --output markdown --all-variations > ~/vault/summary.md

Error recovery

Guardrails

• Don't spam rewrite in a retry loop - each call is a paid AI request.
• If the user didn't name a preset, ask rather than defaulting. "summary" vs "action_points" vs "email" produce different outputs; the wrong choice wastes a call.
• For regeneration of the same preset the user already applied, rewrite will re-run the AI (not cached). Call this out if it seems unintended.
• When presenting the result to the user, show both the preset used and the UUID so they can re-run or reference it.

See vivadicta-cli-usage for exit codes and output formats. See vivadicta-vault-export for piping rewrites into Obsidian.