balaji-dutt/dotfiles/unslop-file-voice

Unslop File Voice

Rewrite a natural-language file with unslop-file and opt in to voice matching. This is separate from unslop-file because it uses personal style memory and should never be the invisible default.

Use this skill when

• The user runs /unslop-file-voice <filepath>.
• The user asks to humanize a file and match their saved voice profile.
• The user provides --voice-sample <path> for a one-off file rewrite.

Do not use this skill when

• The user asks for deterministic/no-network cleanup. Voice matching is LLM-only.
• The file is code, config, secrets, credentials, lockfiles, or any sensitive path.
• No saved voice profile or explicit voice sample is available. Ask the user to create/provide one instead of doing a generic rewrite.

Voice source order

1. If the user supplied --voice-sample <path>, use that sample for this run.
2. Otherwise look for a saved numeric style profile without printing its contents:

• $UNSLOP_STYLE_MEMORY
• $XDG_CONFIG_HOME/unslop/style-memory.json
• ~/.config/unslop/style-memory.json
• %APPDATA%/unslop/style-memory.json on Windows

1. If no profile exists, stop and tell the user to run:

unslop --save-voice-profile samples/my-writing.md

The saved profile stores numeric style targets only. It does not store prose samples.

Process

Prefer the scripts/ directory from the sibling unslop-file skill.

Common layouts:

• skills/unslop-file-voice/SKILL.md plus skills/unslop-file/scripts/
• ~/.claude/skills/unslop-file-voice/SKILL.md plus ~/.claude/skills/unslop-file/scripts/
• Installed CLI fallback: unslop

Steps:

1. Confirm the target is .md, .markdown, .txt, .rst, or extensionless prose.
2. Refuse sensitive files and non-prose/code/config formats using the same boundaries as unslop-file.
3. Resolve the voice source. Do not print profile JSON or sample text.
4. If using saved memory, fail closed with a preflight load check before rewriting. From the unslop-file skill directory, run:

python3 - <<'PY'
from scripts.style_memory import StyleMemoryError, load_profile
 
try:
    profile = load_profile()
except StyleMemoryError:
    profile = None
if profile is None:
    raise SystemExit(
        "no unslop style memory found; run "
        "unslop --save-voice-profile samples/my-writing.md first"
    )
PY

If this check exits non-zero, stop. Do not run --voice-memory, because the CLI treats missing memory as a generic rewrite fallback.

1. Run LLM mode with voice matching:

python3 -m scripts --voice-memory --mode full <absolute_filepath>

For a one-off sample:

python3 -m scripts --voice-sample <sample_path> --mode full <absolute_filepath>

If only an installed CLI is available, use the same flags with unslop.

1. Preserve fenced code, inline code, URLs, paths, commands, headings, tables, YAML frontmatter, numbers, versions, and errors exactly.
2. Report the rewritten file path, backup path, mode, and which voice source type was used: saved profile or explicit sample. Do not reveal profile contents.

Defaults

• Default intensity: --mode full.
• If the user explicitly asks for subtle or balanced, pass that --mode instead.
• Do not pass --deterministic; it ignores voice matching.
• Do not use exported markdown profiles (unslop-voice-profile.md) for the CLI run. The CLI expects saved JSON style memory or a prose sample path.

Boundaries

• This skill changes files in place and writes FILE.original.md backup first.
• If validation fails, report the error and leave the original restored per unslop-file behavior.
• Voice match means cadence, register, punctuation, contraction rate, and sentence rhythm. It does not mean inventing facts, preferences, biography, opinions, or emotional tone.