A Claude Code skill that produces writing with zero em dashes, and, just as important, writing built out of sentences that never reach for one in the first place. Heavy em-dash use has become one of the most recognizable fingerprints of machine-written text.
Dex is the agent-native analytics engineering toolkit. Point it at your warehouse and your dbt project. It learns the landscape, authors your transformations, and tells you exactly what to fix when the schema drifts. Built for analytics engineers and data engineers who want more out of their coding agent.
@exmergoFAQ
no-em-dashes is a Claude Code plugin with 1 hand-picked skill for content work, indexed on Flowy. Install it with the command on its page. It includes no-em-dashes. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
$ npx -y skills add exmergo/skill-no-em-dashes --agent claude-code
A Claude Code skill that produces writing with zero em dashes, and, just as important, writing built out of sentences that never reach for one in the first place.
Heavy em-dash use has become one of the most recognizable fingerprints of machine-written text. Many readers now skim a dash-laden paragraph and think "a bot wrote this." This skill keeps prose reading as a person's, and it forces tighter, more deliberate sentences along the way.
Built and shared in the open by Exmergo, the team building AI agents for data analytics. See more of our work at exmergo.com/open-research.
This skill is published through the Exmergo marketplace. Add the catalog once, then install the skill. Inside Claude Code:
/plugin marketplace add exmergo/exmergo-agent-plugins
/plugin install no-em-dashes@exmergo
The skill then loads automatically and runs whenever Claude writes or edits prose for
you. To update later, run /plugin marketplace update exmergo.
If you would rather not use the plugin system, copy the skill folder into your skills directory:
git clone https://github.com/exmergo/skill-no-em-dashes
cp -r skill-no-em-dashes/skills/no-em-dashes ~/.claude/skills/ # personal, every project
# or, project-scoped:
cp -r skill-no-em-dashes/skills/no-em-dashes .claude/skills/
Claude Code auto-discovers any folder containing a SKILL.md. Either way, you do not
have to mention em dashes; clean output is the default.
The two-command install above is specific to Claude Code, which has a plugin marketplace. Most other agents do not install packages; they read committed instruction files. The good news is that one file reaches almost all of them.
AGENTS.mdThis repo ships an AGENTS.md, the open standard (now under the Linux
Foundation) that Cursor, GitHub Copilot agent mode, Codex, Continue, Aider, Windsurf,
and others read automatically. To get the behavior in your own project, copy the
contents of our AGENTS.md into your repository's AGENTS.md (merge if you already
have one). That single step covers the majority of agents, with no install command.
AGENTS.md is the canonical source of the guidance. The per-tool files below are thin
pointers to it, so nothing drifts.
| Agent | File in this repo | How to adopt |
|---|---|---|
| Cursor | .cursor/rules/no-em-dashes.mdc | Copy into your repo's .cursor/rules/, or use Settings -> Rules -> Remote Rule (GitHub) and paste this repo's URL to auto-sync. |
| GitHub Copilot | .github/copilot-instructions.md | Copy into your repo's .github/. |
| Windsurf | .windsurf/rules/no-em-dashes.md | Copy into your repo's .windsurf/rules/. |
check_em_dashes.py scan only helps in agents that can run a terminal command (Cursor
agent mode, Windsurf, Claude Code), and only when the rule tells them to run it.When the skill is active, any prose Claude writes or edits comes out free of:
U+2014),word--word or --), a hyphen padded with
spaces (-), a stray en dash between words, and the horizontal bar.The point is not to swap one dash character for a look-alike. It is to recognize the sentence shape that wanted the dash, then rewrite with real punctuation: a comma, semicolon, colon, period, or parentheses. The skill includes a playbook for each job an em dash typically does (the dramatic reveal, the "not just X but Y" pivot, the interruptive aside, joined clauses, the summary dash, and interrupted dialogue) with worked before-and-after examples.
It deliberately leaves legitimate hyphens alone: compounds like well-known and
state-of-the-art, number and date ranges like 10-20 and 2019-2024, and hyphens
in code, file paths, URLs, and math.
AGENTS.md canonical cross-agent guidance (see "Use with other coding agents")
.claude-plugin/
plugin.json Claude Code plugin manifest (name, version, author)
.cursor/rules/no-em-dashes.mdc Cursor rule, points to AGENTS.md
.github/copilot-instructions.md GitHub Copilot instructions, point to AGENTS.md
.windsurf/rules/no-em-dashes.md Windsurf rule, points to AGENTS.md
skills/
no-em-dashes/ the Claude Code skill itself
SKILL.md the instructions Claude Code loads
scripts/check_em_dashes.py a checker that flags em dashes and stand-ins
evals/ test prompts (evals.json) and recorded results
results/em_dash_counts.csv em dash counts per eval, with skill vs without
tests/ standard-library unit tests for the checker
MAINTAINING.md release flow, private-to-public, and sync notes
CONTRIBUTING.md CODE_OF_CONDUCT.md LICENSE
The writing guidance lives in two self-contained files on purpose: SKILL.md for
Claude Code and AGENTS.md for every other agent. They mirror the same rule because
each ships independently; see MAINTAINING.md for how they stay in sync.
The bundled checker is a single dependency-free Python script. Use it on a file or through standard input:
python skills/no-em-dashes/scripts/check_em_dashes.py path/to/draft.txt
echo "Your text here" | python skills/no-em-dashes/scripts/check_em_dashes.py
It prints every hit with its line, column, and a suggested fix, and it exits with a nonzero status if anything is found, so you can wire it into a pre-commit hook or CI step. It skips fenced code blocks, inline code, and URLs so it does not flag the hyphens that belong there.
Yes. We ran five realistic prompts twice each, once with the skill active and once with a skill-free baseline, then counted the em dashes in every output.
A note on method: the baseline runs are generated with the skill fully disabled. A skill that is installed globally will otherwise clean the baseline too and hide the effect, so isolating it matters.
| Prompt | Length | Baseline | With skill |
|---|---|---|---|
| Marketing launch copy | ~150 words | 0 | 0 |
| Team email | ~250 words | 3 | 0 |
| Interrupted dialogue | ~200 words | 5 | 0 |
| Essay | ~1,000 words | 6 | 0 |
| Landing page | ~1,000 words | 18 | 0 |
| Total | 32 | 0 |
The skill held every output at zero. The skill-free baseline produced em dashes in four of the five prompts, and the count climbed with length and rhetorical heat, peaking at 18 in the landing page. Short, punchy copy was the only baseline that stayed clean, which fits the pattern: em dashes surface most in longer, reflective, or persuasive prose.
The prompts live in evals.json, the per-eval
counts in em_dash_counts.csv,
and example outputs under
evals/results/examples/. You can also
explore the results in an interactive chart on Exmergo Viz,
our AI agent for charting and dashboards.
The checker has a standard-library test suite (no dependencies):
python -m unittest discover tests
Contributions are welcome. See CONTRIBUTING.md for how to propose changes, and please keep your own prose and commit messages em-dash free (the checker will tell you if you slip).
The skill collects no data and runs locally. See PRIVACY.md.
MIT.
.claude-plugin/
plugin.json
.cursor/
rules/
no-em-dashes.mdc
.github/
copilot-instructions.md
workflows/
ci.yml
.gitignore
.windsurf/
rules/
no-em-dashes.md
AGENTS.md
CHANGELOG.md
CODE_OF_CONDUCT.md
CONTRIBUTING.md
LICENSE
MAINTAINING.md
PRIVACY.md
README.md
skills/
no-em-dashes/
evals/
evals.json
results/
benchmark.md
em_dash_counts.csv
examples/
breakroom-dialogue.baseline.txt
breakroom-dialogue.with-skill.txt
release-delay-email.baseline.txt
release-delay-email.with-skill.txt
README.md
scripts/
check_em_dashes.py
SKILL.md
tests/
test_check_em_dashes.py© 2026 Flowy · Free and open source
Built for Claude Code · Not affiliated with Anthropic