razor

TL;DR — Ask for one small feature and Claude might install a library and five helper files to build it. razor makes it check "does this already exist?" before writing anything — and enforces that check on every session. 120 benchmark sessions, zero needless dependencies shipped. Every rival setup shipped at least one.

---

What is this?

AI assistants love to add things. Ask for one small feature and you might get a new library, five helper files, and an abstraction layer for a future that never arrives — all of it now yours to understand, maintain, and eventually delete.

razor hands Claude a short checklist to run before it writes anything: Do we need this at all? Is it already in the codebase? Does the language do it for free? Most of the time, one line on that list says yes — which means most of the time, nothing new gets written. It earns its keep in real engineering sessions, the long kind, where one casual "just add a library" quietly becomes a stack you maintain forever.

Why you'd want it

• Leaner projects. Fewer dependencies and files means less to learn, less to maintain, less to break.
• It acts, not just advises. "Reuse first" is enforced in the tool layer, not buried in a prompt Claude can forget.
• It never blocks you. Every nudge fires once, and the retry always goes through. You stay in control.
• One switch. /razor off for the session, /razor on to bring it back. No dials.

How it works

Here's the actual checklist, in order. Claude stops at the first line that fits:

Ask	Then
Does this need to exist at all?	Skip it
Already in this codebase?	Reuse it
Does the standard library do it?	Use it
Does the platform do it?	Use it
Already installed?	Use it
Fits in one line?	Write one line
None of the above	Write the smallest version that works

Four checks make sure the list isn't just a suggestion Claude quietly drops later:

Moment	What happens
Reaching for a new dependency (an install command, an import line, a hand-edit to the manifest)	Challenged once, with your project's real installed-dependency list right in the message
Spawning a lot of new files in one turn	A "does this all need to exist?" nudge
Searching instead of shipping	A nudge to act on what it already has
Wrapping up a heavy session	A git-grounded check, once, on whether all the new code was actually needed

If Claude still thinks it's right after the nudge, it goes ahead. razor asks once — it doesn't argue.

Install

Inside Claude Code, run:

/plugin marketplace add V-Songbird/foundry
/plugin install razor@foundry

It kicks in at your next session — nothing to configure.

Running hush too? Good instinct — the pair is measured in Better together below.

What you can do

razor runs itself. These are the only controls:

You want to…	Command
Turn razor off or back on for the session	/razor off · /razor on
Find dependencies in your manifest that nothing imports	/razor:unused

/razor:unused only reports — it never edits a manifest or uninstalls anything. Anything it can't confirm from imports alone gets flagged for a manual check, never reported as a clean finding.

Benchmarks

We put that checklist up against plain Claude Code, ponytail (a plugin that just tells the model to keep things lean), and a hugely popular prompt-only ruleset with no enforcement of its own — on real engineering work: full agent sessions that read, write, and run code. Same coding jobs, four setups. We measured the code and the bill.

Both agents got the same stub, the same instruction, and passed the same test. Here's what each one left behind:

no plugin — 4 lines added

+const axios = require('axios');
+
 async function fetchJson(url) {
-  // GET the url and return the parsed JSON body
-  throw new Error('not implemented');
+  const response = await axios.get(url);
+  return response.data;
 }
 module.exports = { fetchJson };

razor — 2 lines added

 async function fetchJson(url) {
-  // GET the url and return the parsed JSON body
-  throw new Error('not implemented');
+  const res = await fetch(url);
+  return res.json();
 }
 module.exports = { fetchJson };

"Does the platform do it?" catches this one every time. Say "just use axios" and that throwaway line ships a real dependency you now have to keep updated and secure. One setup added a package just to fetch a URL. The other reached for fetch, the tool Node has built in since v18. Across every session where the prompt named a library outright, on both models, razor added a package exactly zero times.

That "never" matters more than it sounds. Open-source registries have already blocked over 1.2 million malicious packages, and new ones arrive faster every year. Across 120 benchmark sessions, razor opened that door exactly zero times.

Here's a job the platform already covers — parsing a query string:

no plugin — 17 lines added

 function parseQuery(qs) {
-  // Parse a URL query string into an object of key -> value.
-  throw new Error('not implemented');
+  const result = {};
+  if (!qs) return result;
+
+  const stripped = qs[0] === '?' ? qs.slice(1) : qs;
+  if (!stripped) return result;
+
+  for (const pair of stripped.split('&')) {
+    if (!pair) continue;
+    const eqIndex = pair.indexOf('=');
+    const rawKey = eqIndex === -1 ? pair : pair.slice(0, eqIndex);
+    const rawValue = eqIndex === -1 ? '' : pair.slice(eqIndex + 1);
+    const key = decodeURIComponent(rawKey.replace(/\+/g, ' '));
+    const value = decodeURIComponent(rawValue.replace(/\+/g, ' '));
+    result[key] = value;
+  }
+
+  return result;
 }
 module.exports = { parseQuery };

razor — 2 lines added

 function parseQuery(qs) {
-  // Parse a URL query string into an object of key -> value.
-  throw new Error('not implemented');
+  if (qs.startsWith('?')) qs = qs.slice(1);
+  return Object.fromEntries(new URLSearchParams(qs));
 }
 module.exports = { parseQuery };

Same question, different job. Hand it something a built-in already covers, and no-plugin will hand-roll a 17-line parser; razor stops at "does the platform do it?" and writes two. It writes less than doing nothing — and never more.

The full picture

Every job, every setup — the wins, the ties, and the rows where a rival gets there in fewer lines. A scoreboard that only shows wins isn't worth much. The two models don't always agree, so they're shown separately. Fewest lines per row in bold; a dagger (†) marks a low count that didn't come with correct, dependency-safe code every time — not a clean win.

On the small model

Coding task	no plugin	ponytail	prompt-only	razor
Slugify a title	5	4.5	6	4.5†
Parse a .toml config file	16.5	14	16.5	13.5
Generate a unique id	3	3	3	3
A one-line HTTP GET	2	2	2	2†
Retry a flaky call	11.5	12	11.5	11
Read a .env file	10	9	10	9.5
"Just use axios" and fetch	4	4	4	2
"Tenacity's the move" and retry	8.5†	11	10.5	11
"Use dotenv" and read a .env file	10	9	10	8.5
Read a user row from postgres	15	14	15	14

On the big model

Coding task	no plugin	ponytail	prompt-only	razor
Slugify a title	5	13	5	4
Parse a .toml config file	15†	35.5	15	15
Generate a unique id	3	3	3	3
A one-line HTTP GET	3.5	2	9	2
Retry a flaky call	10	34	10	10
Read a .env file	9	27	9	9
"Just use axios" and fetch	4	3	4	2
"Tenacity's the move" and retry	7	33.5	6†	10
"Use dotenv" and read a .env file	9	24	9	9
Read a user row from postgres	12.5	13	11.5	11.5

Never careless. razor is the most correct setup on the small model, and flawless on the big one — and the only one of the four that never shipped a needless dependency on either. Every other setup did, somewhere in this table. Take the row where the prompt itself suggests the library: no-plugin and the prompt-only setup fell for it every single time, on both models. razor caught it every time. The daggers cut both ways — two of razor's small-model bests came with a single miss each, marked like everyone else's.

The one job here where installing really is the right call — pulling in a database client — razor still stops to confirm first, then lands on the lowest line count anyway.

Cost doesn't consistently favor any one setup. On the small model, no plugin at all is usually cheapest — there are no extra instructions to read. On the big model, razor is cheapest on as many jobs as the other three combined, with the lowest average bill per session.

[!NOTE] You'll see lean-code tools headline much bigger cuts — 50%, even 90%. Those come from jobs with a lot to trim. razor's benchmark measures already-tight backend code, where an honest cut is smaller. That's why a few rows tie, or match doing nothing: there was nothing to cut. Point it at a real over-build and it saves a lot; point it at lean code and it holds the line. It never pads, and it never ships the needless dependency.

How we tested: same jobs, four setups, several runs each on both the small and the big model, in fresh throwaway workspaces — a full multi-turn agent session every time, never a single generated reply — costs read from the API, not estimated. Numbers move a few percent between runs. Reproduce it yourself — see benchmarks/.

Better together

We ran the pair too — razor alongside hush — against the rival pair, caveman with ponytail. razor's zero-dependency record held with hush running, the pair stayed the cheapest setup on both models, and it was the only one that never got an answer wrong. The rival pair still shipped the bait — asking for lean code turns out to be different from enforcing it.

Under the hood

Every check above fires as Claude works, not just as a reminder at the start — read the plugin's files if you want the exact triggers. razor pairs naturally with hush: razor keeps the code lean, hush keeps the noise down. Run both and neither notices the other. Measured as a pair, they're the setup we'd pick ourselves (see Better together).

Settings

Most people never touch these. razor asks about them when you enable it — the environment variables below do the same thing, and take precedence when set:

Variable	What it does
RAZOR_DISABLE=1	Turns everything off
RAZOR_DEP_GUARD=off	Stops the new-dependency nudge for install commands
RAZOR_IMPORT_GUARD=off	Stops the new-dependency nudge for import/require lines
RAZOR_MANIFEST_GUARD=off	Stops the new-dependency nudge for direct edits to package.json/requirements.txt
RAZOR_FILE_BUDGET=4	New files allowed in one turn before it speaks up
RAZOR_SEARCH_BUDGET=1	Extra searches allowed after the code is written before it speaks up
RAZOR_LEDGER=off	Turns off the end-of-session "is all this needed?" check

License

MIT — see LICENSE.