verve

Is anyone recognisably behind this writing?

A free, open-source tool by DBHQ

verve cuts the AI tells out of a piece of writing and puts a voice back into it, in British English, without changing what it says.

What makes it different

It does both halves. Most de-slopping tools only subtract. Strip the tells and stop, and you get clean prose that still reads as machine-made - just blandly rather than floridly, because nothing is behind it. So the pattern sweep is followed by a voice pass: opinions, rhythm variance, specificity, acknowledged complexity, a bit of mess.

Meaning is a veto, not a trade-off. Every fact, number, name, date and citation survives unchanged; technical terms keep their exact wording; no invented statistics or examples. The result is scored on six dimensions, and fidelity below 9 forces a revision regardless of the total - because that failure is a changed meaning rather than a stylistic one. Aggressive rewriting tempts a model to compress three real points into one punchy line, and that is the failure this guards against.

It knows when to do nothing. Triage comes first. Text that already reads as human-written comes back unchanged with one line saying so. A tool that processes clean prose for the sake of processing it makes writing worse.

It costs nothing to run, and there is no upsell. The whole skill is the conversation itself - no API key, no per-word billing, no network round trip, no paid tier holding the good half back. Earlier versions shipped an optional commercial detector-evasion API; it was removed in July 2026, because writing that reads as human should come from a judgement about the writing, not from a service that tunes text to score well against a classifier.

British English throughout, which is the point if you write for a UK audience and are tired of drafts drifting into American spelling.

Install

As a Claude Code plugin (recommended)

/plugin marketplace add dbhq-uk/marketplace
/plugin install verve@dbhq

Any agent (Cursor, Copilot, Windsurf, Gemini, Cline and more)

npx skills add dbhq-uk/verve-skill

The skills.sh CLI installs into whichever agent directories it finds, so this works outside Claude Code and Codex too.

Local install (Claude Code or Codex)

git clone https://github.com/dbhq-uk/verve-skill.git
cd verve-skill
./install.sh          # Claude Code: symlinks into ~/.claude/skills (edits are live)
./install-codex.sh    # Codex: installs into ~/.codex/skills

install.sh and install-codex.sh are the same install two ways: Claude Code substitutes ${CLAUDE_SKILL_DIR} so the whole skill directory is symlinked untouched, while Codex does not, so its SKILL.md is rewritten at install time.

Nothing to install beyond that. No packages, no virtualenv, no credentials, no network. The skill is instructions and reference material, not tooling.

Usage

Ask in any session. Text comes from the message, a file, or the clipboard, and the options are plain language.

"verve this: [text]"
"verve draft.md"
"verve my clipboard"
"verve draft.md in a casual tone"
"verve draft.md with heavy rewriting"
"verve essay.md and explain what you changed"
"verve draft.md and save to output.md"

Strength is the dial worth knowing. Light touches only the unmistakable tells - banned words, em dashes, chatbot artefacts, sycophancy - and leaves sentence structure alone. Moderate, the default, adds rhythm and voice work. Heavy restructures freely and rewrites most sentences from scratch. The meaning constraints hold at every level, without exception.

How it works

1. Triage - already-human text is returned unchanged rather than mangled
2. Tone - one of four presets, held throughout
3. Pattern sweep - five groups of tells (content, language, style, assistant artefacts, filler), each with before/after
4. Voice pass - put opinions, rhythm and specificity back, inside the tone
5. Quick checks - a 14-item pre-flight list run against the draft
6. Score - six dimensions, fidelity as a veto rather than an average

What this will not do

Route your text through a detector-evasion service. Verve shipped with an optional Undetectable AI engine until July 2026, and it was removed rather than kept as a switch nobody had to flip.

Two reasons, and the second is the real one. It contradicted the free, read-every-line argument the rest of the skill is built on. And it aimed at the wrong target: a classifier score is not the goal, prose a person would be happy to put their name to is. Those come apart the moment you optimise for the score, and when they do, the score wins and the writing loses.

Development

Want to hack on the skill or run it from source with live edits? See docs/dev-setup.md.

CONTRIBUTING.md covers working on it, and AGENTS.md is for an AI agent doing so. The skill itself is skills/verve/SKILL.md.

Acknowledgements

Several patterns - false agency, vague declaratives, narrator-from-a-distance, meta-commentary, emphasis crutches, telling-instead-of-showing - and the idea of a scored exit gate come from stop-slop by Hardik Pandya (MIT).

Renamed from humanize in July 2026. The skill still triggers on "humanise this" and "make this sound human"; only the name and the directory changed.

License

MIT © 2026 DBHQ Consulting Ltd