Flatline scans every skill on your Mac — across Claude Code, Codex, Cursor and OpenClaw — and shows which ones are truncated, conflicting, wasting tokens, or unsafe.
Free & open-source core · Runs 100% locally — your code never leaves your Mac · macOS first
That email doesn't look right. Mind checking it?
Early-bird pricing is locked for you. We'll send one email when early access opens for your wave.
$0 due today. Your reservation locks the early-bird price. We only email you when access opens.
HEALTH SCORE
Needs attention
8 of 32 skills have problems that affect every session.
YOUR ESTIMATE
of your 32 skills never fire
Estimate from community measurements: ~56% of installed skills never invoked; ~300 tokens of description per skill per turn. Your real scan may differ.
deploy-helper instructs the agent to read SSH keys and send them to an external endpoint. Quarantine before it runs.
pdf-tools, db-migrate, figma-export: descriptions cut off when your list overflowed. Installed, but effectively invisible.
code-review vs pr-check overlap 87%. The model picks one at random. That is why reviews feel inconsistent.
Skill descriptions are charged whether they fire or not. 9 skills haven't fired in 30 days; their descriptions alone account for ~410 of those tokens.
Installed to ~/.agents/skills by npx, never linked into ~/.claude/skills. Install said “success”. Claude never saw them.
Sample data shown for illustration. Built on research from Reddit & Xiaohongshu discussions (2026).
of skills were never invoked in community trigger tests. Silent failure: no error, no log.
SOURCE: COMMUNITY TESTS, 2026 RESEARCHinstalled skills had descriptions silently truncated in one real setup. Installed ≠ active.
SOURCE: REAL-WORLD SETUP REPORTED BY USERSof 3,984 scanned skills had at least one security flaw. 13.4% were critical.
SOURCE: SNYK TOXICSKILLS AUDITtokens of skill descriptions ride along every single turn. Used or not, you pay for them.
SOURCE: USER-MEASURED CONTEXT AUDITS// THE SCAN
Installing skills is solved. Knowing what happened after you installed them isn't. Flatline runs the four checks that catch how skills quietly break, across every agent on your machine, in one pass.
See which skills Claude silently dropped when your list overflowed the listing budget. No warning is ever shown. Flatline simulates each agent's budget rules and names the casualties.
Find skills whose descriptions overlap and steal each other's jobs. When two skills fight for the same trigger, the model picks randomly. Flatline shows the pairs and the overlap.
Know exactly what every skill charges you on every single turn, including the ones that never fire. Dead weight becomes a number, not a suspicion.
Prompt injection, hidden Unicode instructions, secret-reading scripts, curl | bash: flagged before they run, not after your keys leave the machine.
The same skill lives in different places for different agents, and installers don't always agree. Flatline maps every location, finds path mismatches, double-nested folders and broken YAML, and tells you how to fix them.
No account. No upload. No config file to write before you see value.
One line via Homebrew. The open-source CLI runs entirely on your Mac.
Flatline finds every skill directory for every agent you use and runs all four checks locally.
A clear, shareable HTML report: health score, every issue ranked by severity, and what to do about each one.
// RESERVE
The scanner is free and open source, forever. You pay only for continuous monitoring and automation, never for finding out what's broken. Pick a plan, drop your email, done.
$0 due today. Reserving locks this plan at this price. We send one email when early access opens for your wave. No card, no commitment.
That email doesn't look right. Mind checking it?
We'll email you when early access opens for your wave.
Team: allowlists, CI gates and audit logs for companies standardizing on skills. On the roadmap; mention it in your reservation email reply if you need it.
No. The scan runs 100% locally on your Mac. Your skills, code and file contents never leave your machine. The report is generated locally as an HTML file you can open or share yourself. The core is open source, so you don't have to take our word for it.
Nothing financial. $0 is due today and no card is taken. A reservation holds your plan at the price shown (early-bird lifetime pricing closes when v1 launches) and puts you in line for an early-access invite, sent in waves. If you change your mind, ignore the email.
Claude Code, Codex, Cursor and OpenClaw at launch, including global and project-level skill directories (~/.claude, ~/.agents, ~/.codex and more). Each agent applies different budget and loading rules; Flatline simulates each one separately.
The core scanner (scanning, all four detection checks and HTML report generation) is free and open source, including the detection-rule library. Pro is a native Mac app (menu-bar monitoring, alerts, one-click fixes) built on that same open core.
Mac first. macOS is where the heaviest agent users are, and it lets us ship one polished experience instead of two half-finished ones. Windows support is on the roadmap. Reserve now and you'll hear when it opens.
Flatline is in development. Early access opens in rolling waves to reservations, first come, first served. Early-bird lifetime pricing is only available during this pre-launch window.
Reserve early access in ten seconds. Be first to run the scan the day it opens.
That email doesn't look right. Mind checking it?
We'll send one email when early access opens for your wave. No spam, no newsletter you didn't ask for.
Which agents do you use? (optional, helps us prioritize)
$0 due today. One email when early access opens. No spam.