terse 1.2.2

Writing rules for Claude Code that lead with the answer and cut mannered prose.

Changes since 1.2.1:

• Listed in the Claude plugin directory. The README install section starts with the listing and the Add to Claude Code button, then the GitHub commands.
• Plugin icon, shipped as a 1024 px PNG so the directory renders it.
• Display name, documentation, support and privacy policy links in the manifest, and a privacy statement in docs/privacy.md: no network calls, nothing collected, one settings key written by the install skill.
• No change to the rules, the reminder, the hooks or the meter.

Install from the directory: https://claude.ai/customize/plugins/id/aa4422b4-a8c3-42e2-8f4f-d505301cb4bd@anthropic-plugin-directory

Install from GitHub:

/plugin marketplace add lowenbjer/claude-terse
/plugin install terse@terse

terse 1.2.1

Writing rules for Claude Code that lead with the answer and cut mannered prose.

Changes since 1.2.0:

• The install skill copies the context meter to the plugin's data directory, which plugin updates leave in place.
• The per-turn reminder says "the Output Style section of your system prompt" in place of a plugin:style label that read as a skill reference in long sessions.
• The install skill is user-invoked only, so its description leaves the per-turn context.
• One-command public-set runner bench/run.py, a model contribution path in docs/models, before/after images.

Install:

/plugin marketplace add lowenbjer/claude-terse
/plugin install terse@terse

Update: claude plugin marketplace update terse then claude plugin update terse@terse, then a new session. Run /terse:install-meter once more to move the meter to the data directory.

terse 1.1.0

Writing rules for Claude Code that lead with the answer and cut mannered prose.

Changes since 1.0.0:

• Rule 19: every sentence has the topic as its subject, no first person.
• Rule 20: a redo sends only the changes.
• Rule 11 drops "i think".
• A UserPromptSubmit hook adds the rule set, condensed to 280 characters, to every prompt against drift in long sessions.
• Opus 5.5 benchmark and a model catalog in docs/models.

Measured on the 12 public prompts, Opus 5.5: first-person hits 1 (vanilla 18, text 1.0.0 7), judge violations per 1k words 1.1 (vanilla 11.1, text 1.0.0 2.7).

Install:

/plugin marketplace add lowenbjer/claude-terse
/plugin install terse@terse

terse 1.0.0

Writing rules for Claude Code that lead with the answer and cut mannered prose. First release: 18 rules as a forced output style, a SubagentStart hook that hands subagents the same text, a context meter for the status line, and the benchmark. Measured on Fable 5.1: 46% fewer words, 51% lower cost on 20 prompts against a real codebase.

Install:

/plugin marketplace add lowenbjer/claude-terse
/plugin install terse@terse

terse 1.2.0

Writing rules for Claude Code that lead with the answer and cut mannered prose. Measured on Fable 5.1 and Opus 5.5: 46% to 54% fewer words and 32% to 51% lower cost on 20 prompts against a real codebase.

Changes since 1.1.0:

• A reply shape at the top of the rules: answer sentence, list or table, at most one caveat, stop.
• Rule 21: bullets start with the fact. No bold lead-ins, no label lines.
• Rule 22: plain words in place of coined names.
• Rule 17: a findings list gets one or two sentences per finding and 300 words.
• Rules with a positive form state it first and the ban second.
• Per-turn reminder rewritten to match, 534 characters.

Measured in a 10-turn conversation on both models: bold lead-ins 20 to 0 on Fable 5.1 and 36 to 0 on Opus 5.5. Tables in docs/benchmark.md.

Install:

/plugin marketplace add lowenbjer/claude-terse
/plugin install terse@terse

Update from 1.1.0: claude plugin marketplace update terse then claude plugin update terse@terse, then a new session.