headroomlabs-ai/headroom (opens in a new tab)
Probably fine
Probably fine. Read the warnings first.
Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
What should I do?
We found no malware tricks, but it misses some signs of a well-run project. Read the warning signs, and download it only from its official GitHub page.
Warning signs 1
Runs a script straight from the internet Good to know
The install step downloads and runs code in one line. Common with real tools too, but you can't see what runs.
Talks to your AI 1
This repo has 1 file that AI coding tools like Claude Code, Cursor and Copilot read as instructions when you open the folder: .github/copilot-instructions.md. We read them and found nothing that tries to take over your AI. It's still worth opening them yourself before you let an AI work in this folder.
Does it cost money?
Its page mentions “subscription”, “enterprise plan”. The code here may be free, but some features or the service it connects to may cost money. Check their pricing page before you rely on it.
Why this grade?
Its checks add up to 86 out of 100, but the score is capped at 84 because its weakest area ("Does it have good security habits?") only gets a C. 70 to 84 is a B. Each area gets its own letter, and the overall grade can be at most one letter better than the weakest safety area ("Is it open about itself?" only costs points). Open a row to see what we checked.
Area grade BDo people trust it?, 23 out of 30 points
Good: Stars. 77,986 people starred it.
Stars are like bookmarks. Malware repos rarely have many real users, and lots of stars means lots of people have looked at it. Stars can be bought, so we also check how fast they came.
Good: Forks. 6,067 copies made by other developers.
A fork is a copy another developer made to build on. Real projects get copied; fake ones rarely do.
So-so: Owner history. Organization created 133 days ago.
Scammers use brand-new accounts that get banned and replaced. An owner who has been around for years has more to lose.
Area grade AIs anyone looking after it?, 22 out of 25 points
Good: Recent updates. Last change 1 days ago.
Software nobody updates stops getting security fixes.
Good: Track record. Repo is 293 days old.
Older projects have had more time for problems to be spotted and reported.
Good: Releases. Publishes versioned releases.
A release is an official, finished version. Without one, you download whatever is in the code that day.
Area grade AIs it open about itself?, 15 out of 15 points
Good: License. Uses the Apache-2.0 license.
A license says whether you're allowed to use it. Serious projects almost always have one.
Good: Instructions. Has a README that explains the project.
A README explains what the project does and how to use it. Malware often has a thin or copied one.
Good: Security contact. Tells people how to report security problems.
A security policy tells people how to report problems privately, so they get fixed before attackers use them.
Area grade CDoes it have good security habits?, 3 out of 5 points
So-so: Known security holes. 7 of the 808 parts it uses have a known security hole (for example block-buffer@0.10.4, paste@1.0.15). Not every hole affects you, but the owner should update them.
Security holes that are public but not fixed yet are what attackers look for first. From OpenSSF Scorecard, or, when it has no data, from the project's list of parts (its lock file) looked up in OSV, Google's free public database of known security holes.
Area grade AAny malware tricks?, 27 out of 30 points
Good: 1 warning sign. Each one is explained under Warning signs.
We look for tricks that keep showing up in malware: requests to turn off antivirus, locked zip files, cracks and free premium, hidden code, and AI instruction files that try to take over your assistant. Serious ones cap the grade, however popular the repo is.
Checked Oct 27, 2026. We re-check this repo every night. A grade is a set of warning signs, not a guarantee. How we score