Cybersecurity senior (minor in Game Design) at High Point University.
I build agent systems that run unattended against real services, and I spend most of my time on what they get wrong quietly.
I find bugs in open-source data infrastructure - job queues, vector stores, agent frameworks - and send the fix upstream with a reproduction and a test that fails on the old code. Many of them are one pattern: a backend fails, and the code above it reports "nothing there" or "done" instead of an error, so the caller carries on as if the data never existed.
47 merged across 19 projects. See them on GitHub (46 there; the cognee fix was merged through the maintainer's own pull request, #5320).
| Project | What was wrong |
|---|---|
| coverage.py #2339 | coverage combine could treat a re-saved parallel data file as a duplicate and delete it, losing that process's later data (released in coverage 7.17.0) |
| Celery kombu #2767 | A connection error while putting an unacked Redis message back deleted the message instead of restoring it |
| Haystack #4006 | In pgvector, delete_by_filter with an empty filter deleted every document |
| semantic-router #706 | A failed Qdrant read looked like an empty index, so a "remote" sync then deleted the user's local routes |
| pgqueuer #996 | An empty entrypoint list made clear_queue and the log clears wipe the whole table instead of nothing |
| langgraph-redis #213 | The shallow savers returned pending writes from earlier checkpoints, so a second resume with an update could fail (a user hit exactly this in #215; I added the regression test the maintainer asked for in #219) |
| rq #2552 | Retry intervals given as a generator were lost, so on the first retry the worker quit and the job was left stuck as started |
| tox #4203 | With constrain_package_deps = true, tox copied extras, editables and local paths into its constraints file, which pip rejects, so an ordinary deps = coverage[toml]>=7 failed to install |
By project
pgqueuer 12, langgraph-redis 6, rq 6, Feast 3, Haystack 3, semantic-router 2, Semantic Kernel 2, tox 2, Agno, cognee, coverage.py, Celery kombu, LangChain AWS, LangChain Azure, langroid, opik, redis-py, RedisVL, mcp-server-manager
Fixes open across LlamaIndex, Semantic Kernel, crewAI, Dify, LangChain and others. See them all.
If your project wraps a database, vector store, cache or queue, I will scan its adapters for this defect for free and send you a count and one finding worked in full. Fixes are a fixed price agreed before I start, and every one ships with a test that fails on your current code. eric.delorefice@outlook.com
unattended-agent-security - ten failures from running an LLM agent unattended against live systems, and the controls that prevent them. Nine of the ten reported success while failing.
A check that has never gone red proves nothing. Before I trust a fix, I make its test fail against the old code and read why it failed.
- Microsoft Certified: Azure Fundamentals (passed 30 September 2026)
- Interested in agent reliability, and the gap between a system that reports success and one that succeeded