CES hooks into Claude Code at the process level. Every session runs 177 checks. Exit only on green. $49 one-time.
Payments processed by Stripe. 7-day false-done guarantee or full refund.
Spots remaining: 93 of 100 — updated weekly
"All tests pass.
Here’s how to verify
it yourself."FAIL: test_auth_token
AssertionError: token
expired — not checkedThat sentence has cost you more hours than any bug. The loop ran 38 iterations. Claude exited on iteration 12. The remaining 26 ran against a broken state you discovered manually.
"You can verify this
by running:
npm run test"47 passed, 0 failed
Build: success
Commit: readyIt doesn’t ship work. It ships instructions for you to do work. “You can verify this by running…” is not an output. It’s a handoff. The token cost is indistinguishable from a session that actually shipped.
Iteration 38/???
Claude: "Still working
on this…"
[billable time burns]Iteration 3
[CES] EXIT_SIGNAL: true
47 passed, 0 failed
Session complete.Iteration 38 of infinity. Without an enforced exit condition, Claude iterates forever or exits prematurely. Both failures cost the same — billable time, no output.
“I wasted about 5 hours today trying to accomplish tasks that could have been done in 30–40 minutes… Beyond the usual infinite loops Claude Code often finds itself in (it has been executing a simple file refactor task for 783 seconds as I write this), the 4.0 models have the fun new feature of consistently lying to you in order to speed along development.”
“Claude is like an extremely confident junior dev with extreme amnesia, losing track of what they’re doing easily.”
“The other day we spent four and a half hours trying to fix something. Going in circles. Finally I said: start over from scratch. It picked a different approach and everything worked. That happens every week.”
“Anyone else here doing full-stack Next.js in Cursor and watching the Claude quota evaporate before lunch? Massive context windows from all the components, pages, and DB logic would smoke the default limits fast.”
“Claude Code was spending 85% of its context window reading node_modules… 85,000 out of 100,000 tokens were being consumed by dependency code, build artifacts, and git internals.”
Claude Code has a hook system — small scripts that run at specific points in every session. CES installs three hooks:
claude-loop) — wraps your entire session. Dispatches Claude, monitors exits, only terminates when the Stop hook returns green.No changes to your workflow. No new IDE. No subscription.
pip install ces && ces installbrew install cesWorks on macOS and Linux. Windows: see docs.
False exits are common — see Reddit threads above for real accounts.
Every edge case that causes false done-claims has a named test. The full test file is public. Read it before you buy.
Get stress_test.py — emailed within 1 hour →90-second animated walkthrough of CES intercepting a false exit. Real hook mechanics, real terminal output, real test failures caught.
Watch the demo above →CES is a Python wrapper. The hooks are visible. Every file is readable before you commit $49.
Email for source access (full repo publishing soon — guaranteed) →claude-loop wraps your Claude Code session. You run one command instead of watching the terminal.
After every Claude exit attempt, CES runs your test command (pytest, npm test, any shell command). If it fails, Claude cannot exit.
When tests pass, EXIT_SIGNAL: true is confirmed. Session terminates. You get a commit, not a checklist.
No test command? CES still runs — it uses EXIT_SIGNAL pattern matching to detect genuine completion vs. premature exits.
Founding price locks at $49. After 100 seats: $99 permanently.
Spots remaining: 93 of 100 — updated weekly
One invoice. One install guide.
Volume discount: 10 seats = $390 vs $490 solo. One purchase, one receipt.
“If CES doesn’t catch a single Claude false-done in your first 7 days of real use, email us. Full refund, no questions, and you keep the tool.”
Payments securely processed by Stripe. We never see your card details.
No. Prompts live in Claude’s context window — Claude can reason past them. CES is a Python wrapper with process-level hooks. The Stop hook runs at the shell layer. Claude doesn’t “ignore” a shell interrupt.
Claude Code exposes a hook system — shell scripts that execute at defined points in a session lifecycle. The Stop hook fires every time Claude attempts to end a session. CES registers a Stop hook that runs your test command. If tests fail, the hook returns a non-zero exit code and Claude continues. This is not a prompt instruction — it is a process interrupt.
CES still works. Without --test-cmd, CES monitors for EXIT_SIGNAL: true in Claude’s output — a verified completion pattern. Claude is prompted to emit this only when it has genuinely verified its work. It’s less rigorous than a real test suite but catches the majority of premature exits. For production use: add pytest, npm test, or any shell command that exits 0 on success.
No. CES hooks into the Claude Code CLI specifically. Cursor Agent (Composer) is a separate surface — CES cannot intercept its exit signals. If you use Cursor Agent, CES is not the right tool today. We’re tracking demand for Cursor Agent support. Join the Cursor waitlist →
One ~/.claude/ config installation per developer machine. Each dev runs their own binary. One license per seat. The shared CLAUDE.md template (included in team tier) means one config file covers your team’s standards — each dev installs it in their own environment. For 4 devs: 4 licenses, one invoice if you use the Team card.
Price goes to $99 — permanently. Same package, same updates, same guarantee. No tiers, no subscriptions. The founding distinction is only about price.
We refund it. No form, no questions. The only reason to hesitate is if you believe your Claude Code sessions never produce false done-claims. If you’ve been using Claude Code for more than two weeks, you already know that’s not true.
pip install ces && ces install takes two minutes. After that, every Claude Code session runs through the hook system automatically. No new IDE. No new interface. No change to your prompting workflow.
In our internal testing, the agent caught a misleading 100/100 audit before it reached production. The audit claimed all checks passed; the Stop hook ran the actual test suite, found 3 failures, and forced another iteration. Without CES, that broken state would have shipped. A developer running a similar pattern manually described spending 15 hours recovering from a false-done chain that CES would have intercepted on iteration 2.
One false-done catch is worth more than $49. If it doesn’t happen in week one, you don’t owe us anything.
Claim Founding Seat — $49 one-timeFounding price: Spots remaining: 93 of 100 — updated weekly
After seat 100: $99 permanently. No extension.
CES currently hooks into Claude Code CLI only. Cursor Agent (Composer) support is in planning. Leave your email and we’ll notify you when it ships.
Join Cursor Agent waitlist →