Show and tell: RetryScope — find where your Claude Code agent got stuck (local)

[flower1201]

I kept losing time when Claude Code repeated the same failing tool call. RetryScope is a local-first CLI: point it at a session JSONL, get an offline HTML report with evidence-linked “repeated failure” findings and a clickable timeline. No cloud upload. No LLM API.

Pre-release (not on npm): https://github.com/flower1201/retryscope

git clone https://github.com/flower1201/retryscope.git
cd retryscope && npm install && npm run demo
# open examples/out-demo/index.html — expect RF-001

Limits: findings are potential patterns, not token-waste accounting; tool “success” ≠ user task solved; reports omit prompts/commands by design.

Looking for Claude Code users to try the synthetic demo and say whether the report helps. Feedback welcome in Issues / this Discussion — stars not required.

---

做了个本地 CLI：读取 Claude Code 的 session JSONL，检测「同一工具反复失败」类模式，输出离线 HTML（时间线 + 证据）。不上云、不调 LLM。本人作者，欢迎试用反馈。