M.S. in Computer Science @ CQUPT (2027). I build Agent Harnesses — the runtime layer that decides whether an AI agent's output can be trusted. I verify my work with reproducible benchmarks, adversarial boundary probing, and upstream contributions, not adjectives.

My design philosophy, shared across everything I build:

Let the model think. Let the program decide. LLM output is a claim; only a tool result is evidence.

The corollary I keep returning to: capability benchmarks measure the model, but the harness is what makes a result trustworthy. So I work on the layers between "the model said something" and "the system can act on it" — constraint boundaries, evidence gates, and evaluation you can reproduce.

An open-source runtime for running, constraining, and evaluating AI coding agents. The core contribution is architectural: treating constraint and verification as first-class layers that sit between reasoning and tool execution, rather than as afterthoughts.

Why it matters: Most agent projects prove capability with benchmarks, then ship a different system to users. Xenon's interactive mode and evaluation harness share the same isolation boundaries and verification gates — so the 40.0% SWE-bench_Lite pass rate is a valid claim about the CLI you'd actually use.

Key results:

• SWE-bench_Lite: 40.0% instance-level (12/30, single attempt) on deepseek-v4-flash — +6.7pp over the previous version in same-model A/B. Full methodology, sampling seed, and cost accounting are committed and reproducible (report).
• Security hardening (v0.8.5): Found and fixed two real sandbox escapes via systematic boundary probing — path fence symlink bypass + runtime binding gap. Each came with POC, validity self-check (revert → must fail), and regression tests. Verified in production use.
• Architecture: 7 swappable reasoning paradigms, Evidence Runtime verification layer, MCP-native tooling, 2300+ unit tests with CI on every commit.

The project demonstrates that trustworthiness is an architectural property, not a tuning target.

A code-diagnosis workbench that splits responsibility deliberately: language frontends and deterministic analyzers own source facts; the LLM may only propose hypotheses; resolvers and validators decide whether a claim binds back to real operations, types, and control flow. Under-supported conclusions stay unknown/abstained — they are never promoted to findings.

• Real upstream bugs found and verified before submission — e.g. stunner #89 (TURN connection leak) confirmed fixed upstream; issues/PRs to sniproxy, Robyn, qscan.

Merged into DeepSeek-Reasonix (DeepSeek's official agent framework):

• #7086 — preserve advisory diagnostics without weakening the evidence gate (merged 2026-07-31)
• #6618 — honour finish_reason=stop on reasoning-only final answers (merged 2026-07-17)

Open / under discussion:

• NousResearch/hermes-agent #80155 — gate finalization on active todos; surfaced a boundary that drew maintainer triage and a "facing the same problem" from the community.
• volcengine/veadk-python #716 — Xenon Skill + MCP interoperability docs.
• deepseek-ai/awesome-deepseek-agent #301 — listing Xenon as a DeepSeek V4 terminal agent.

Research interests:

• Agent harness design: constraint boundaries, evidence gates, and trust semantics
• Reasoning-paradigm comparison and evaluation methodology
• Static analysis + semantic IR for deterministic code understanding
• Cost-efficiency in production LLM applications

Currently exploring:

• Cross-language semantic analysis (Python, Go, TypeScript, Rust)
• MCP protocol integration for agent tool ecosystems
• Benchmark design for agent reliability measurement

Seeking: AI Agent / LLM Infrastructure engineering roles (2027 grad, available from July 2027)

What I bring:

• Reproducible evaluation methodology (SWE-bench official harness, same-model A/B, committed seeds)
• Adversarial self-review: I probe my own systems for boundary failures and fix them with POCs and regression locks
• Architecture that keeps guarantees consistent between benchmark and production paths
• Cross-stack engineering (Python/Go/C++/TypeScript, static analysis, distributed systems)

• GitHub: @xianyu-sheng
• Email: xianyu.sheng.research@gmail.com
• Technical blog: 掘金 @闲余生
• Location: Chongqing / Chengdu, China (open to relocation)

_{💡 Open-source philosophy: Every claim should be reproducible. Every benchmark should publish its methodology. Every "smart" system should show its work.}