Free LLM Tokens Are Not Free Trust: A Pairing Session That Ended with a Boundary

说明与用途

在免费 LLM 端点上进行探测，发现了部分写入、隐藏状态和模式不匹配的风险。通过实施五个检查，使可抛弃的 fixture 生成工作变得安全。

问题

Using a free LLM tier for a nightly fixture‑generation job risked partial writes and silent failures when the token grant ran out mid‑run.

结果

All five checks passed, confirming the free tier can safely run the disposable fixture job without risking partial data or hidden state.

工作流

• 1. Ask if token grant can run out mid‑run and what happens.
• 2. Define acceptance criteria from questions.
• 3. Write a single‑file probe script with five checks.
• 4. Run probe against the free endpoint before migration.
• 5. If all checks pass, move the fixture generator to MonkeyCode.
• 6. Monitor for failures and re‑run probe if changes occur.

证据

The probe sends five short requests, averages the reported tokens, and projects requests against the advertised grant.

来源

The probe posts a canary string, opens a new conversation, and fails if the canary appears in the second reply.

来源