More screenshots
Works with the AI Agents you already use
LoopBreaker MCP — Anti-Death-Spiral & Test Grader
Out-of-band AST test grader, reward-hacking firewall & anti-death-spiral engine. Prevents AI agents from burning $15–$30 in circular loops and mutating test assertions.
$9
LoopBreaker MCP — Anti-Death-Spiral & Test Grader
Example session with this skill installed
Run the test suite on /src/services/billing.ts using LoopBreaker. Protect the test suite from assertion tampering, detect any circular debugging loops, and fix the failing invoice currency issue.
- Read your context and instructions
- Compiled the loopbreaker mcp
I have executed LoopBreaker MCP across the billing service test suite. The cryptographic vault prevented test assertion tampering, broke a 3-turn circular editing spiral, and applied a verified AST patch.
LoopBreaker Defense & Verification Report
<cryptographic_vault status="LOCKED">
<suite_hash algorithm="sha256">7f01a89bc42e91a03f88d401...</suite_hash>
<tampering_attempts detected="1" action="BLOCKED_EXIT_CODE_2">
Attempted mutation on tests/billing.test.ts:38 (removal of expect(res.status).toBe(200))
</tampering_attempts>
<baseline_integrity verified="true" />
</cryptographic_vault>
Circular Loop & Reduced Steer Diagnostics
| Cycle | File Target | Edit Signature | Cost Incurred | Loop Action |
|---|---|---|---|---|
| Turn 1 | src/billing.ts | Add optional chaining ?.code | $0.42 | Tracked |
| Turn 2 | src/billing.ts | Revert to default parameter currency = 'USD' | $0.45 | Tracked |
| Turn 3 | src/billing.ts | Re-apply ?.code (A ➔ B ➔ A Pattern) | $0.41 | INTERCEPTED (Break Spiral) |
// Verified Surgical Fix: src/services/billing.ts
export function formatInvoice(data: InvoiceData): FormattedInvoice {
// Resolved root cause: currency object was nested inside metadata payload
const currencyCode = data.metadata?.currency?.code ?? data.currency ?? 'USD';
return {
invoiceId: data.id,
currency: currencyCode.toUpperCase(),
totalFormatted: formatCurrency(data.amount, currencyCode)
};
}
Next Steps
- The test suite passed with 100% assertions green under cryptographic verification.
- Verified that zero test assertions were altered or bypassed.
- Ready for staging deployment and PR creation.
Connects securely to your tools. The creator never sees your data.
About this skill
LoopBreaker is a local runtime enforcement engine designed to eliminate the two most expensive failure modes in autonomous coding agents:
-
The Infinite Debugging Death Spiral: Repetitive circular edits (A ➔ B ➔ A) that burn $15 to $30 in token costs without resolving test failures.
-
Reward Hacking & Test Assertion Corruption: Agents modifying test files, disabling assertions, or altering mock expectations instead of fixing the application code.
-
Cryptographic Test Suite Vault:
- Takes a SHA-256 snapshot of all test suites before agent execution.
- Enforces a zero-mutation policy on test assertions via PreToolUse hooks (Exit Code 2).
-
Deterministic Loop Breaker:
- Maintains an in-memory hash ring of recent file edits.
- Detects cyclical permutations across turns.
- Forcefully breaks the loop when a cycle is identified.
-
Reduced Steer Protocol:
- Compresses 4,000+ line terminal stack traces down to a concise ~200-token packet:
[PINNED GOAL] + [VERBATIM FAILURE DELTA]. - Prevents context window dilution and saves up to 95% of debugging tokens.
- Compresses 4,000+ line terminal stack traces down to a concise ~200-token packet:
grade_test_run: Executes test suites through the out-of-band grader and returns a Reduced Steer payload.detect_edit_loop: Audits recent file diffs against the hash ring to catch circular edits.lock_test_suite: Generates cryptographic locks for test directories.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
Requires tests to be inspectable locally via Node or Jest/Vitest/Pytest. Operates on local worktrees and file systems; remote cloud CI loops must run the pre-tool hook in their runner environment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 19 days ago
- Software
Node.jsterminal
Git
Permissions
No special permissions declared or detected