Free · deterministic
Shorter prompts. Nothing lost.
Compaction cuts the filler, repeats, and padding out of prompts and agent instructions so agents read fewer tokens — and never touches code, quotes, numbers, or a single “never” or “only”.
Before
Please note that you need to run the tests prior to merging. In order to do this, make sure that you utilize the CI pipeline. Do not force push to main.
After
Run the tests before merging. To do this, ensure you use the CI pipeline. Do not force push to main.
−34.2%
same instructions, same “Do not”
Paste a prompt. See what an agent no longer has to read.
This is the same engine the workspace, API, MCP server, and CLI run. Safe keeps your wording and only removes what carries no instruction; Aggressive also drops intensifiers and articles.
−28.3% · ~29 tokens saved · 410 → 294 chars
Runs entirely in your browser — nothing you paste is sent anywhere.
Removes padding. Keeps every instruction.
Compaction is a fixed set of rules, not a model rewriting your prompt. The same input always gives the same output, and running it twice changes nothing more.
What it removes
Politeness and hedges, second-person scaffolding (“You need to…”), verbose phrases, repeated lines, and whitespace or table padding. Aggressive also drops intensifiers and articles.
What it never touches
Code blocks, inline code, front matter, headings, quoted text, URLs, file paths, flags, and template placeholders stay byte-for-byte.
A guard on every edit
Any edit that would drop a negation or limiter (not, never, only, unless, must), a number, or an identifier is skipped. Nothing is removed right after a negation.
Measured in tokens
Every shorter word or phrase is checked against a real tokenizer, so a swap that looks shorter but costs more tokens never ships.
Bring Rubrkit into your toolchain.
Sync graded artifacts into the repos where your agents run, or connect any MCP client to the same audits, bundles, and compaction tools the web app uses.
CLI
Pull approved artifact bundles into local projects and place them where Codex, Claude, or generic agents expect their instructions.
# Pull approved artifacts into your project npx rubrkit pull # Place them where your agent expects them npx rubrkit pull all --agent claude --yes
MCP server
Point any MCP client at Rubrkit and call the same artifact bundle, audit, and compaction tools, authenticated with your Rubrkit API key.
{
"mcpServers": {
"rubrkit": {
"url": "https://www.rubrkit.com/api/v1/mcp",
"headers": {
"Authorization": "Bearer <your-rubrkit-api-key>"
}
}
}
}Answers before you start.
Know which instructions are ready to run.
Compact in your workspaceFollow the review loop as it ships.
Notes on AI artifact testing, prompt compaction, evals, and proof reports.