Lindy Xu lindixu6-hash

Production AI Agent Engineering

Full-stack AI Agent developer building executable evals, tool-security boundaries, context systems, and human-gated workflows.

全栈 AI Agent 开发者，专注可执行评估、工具安全、上下文工程与人在回路工作流。

Not another resource list: an executable production-readiness gate for AI agents, maintained in English and Simplified Chinese.

不是另一份资源清单： 一套可执行、可审查、完整中英文同步的 AI Agent 生产就绪门禁。

Current evidence / 当前证据	Verified result / 已验证结果
Release	`v0.19.1`
Deterministic tests	`156` total: `153` pass, `0` fail, `3` optional integrations skipped; CrewAI pytest `4/4`
Prompt-injection fixtures	`8/8` across reference, LangGraph.js, OpenAI Agents SDK, and CrewAI delegated-task runtimes
CrewAI delegation	Attested Release evidence proves real Agent + Task.context + Crew.kickoff(), 8 cases x 5 evidence files, and zero network attempts; the deterministic bundle is a durable Release asset, not a benchmark or safety certification
Provenance	Separate producer + SHA-pinned verifier, authority-policy digests, GitHub OIDC/Sigstore attestation
Starter supply chain	Generated workflows pin `actions/checkout` to the reviewed v7.0.1 commit SHA
Governance	CFF 1.2 citation, CODEOWNERS, funding boundaries, protected `main`
Distribution	Automated v0.19.1 release run checksums and attests the npm tarball, SBOM, and CrewAI evidence bundle, then a separate write-scoped job publishes and byte-verifies all four Release assets; npm publication remains explicitly pending

Try / 使用	Integrate / 接入	Inspect evidence / 查看证据	中文
Web scorecard	Five-minute fail-closed setup	Attested eval provenance	中文 README
OpenAI Agents SDK eval	Public JSON Schemas	Source-linked incidents	中文 Schema 指南

AI Content Workflow Skills is a same-maintainer consumer. Its score gate passes at 12/20, while its draft-only profile and three launch blockers remain visibly failing.
PaperSage #149 is an independent Agent Card adoption PR under maintainer review. It is not counted as adoption unless merged.

mac-developer-bridge #4: a poisoned-repository adversarial harness merged after the upstream owner reproduced the evidence and CI passed. This is a merged upstream contribution, not adoption or endorsement of Awesome Agentic Engineering.

Invited as a Write collaborator for tickernelz/opencode-mem, a 1.3k+ Star local-first memory plugin for coding agents.
Independently reviewed, verified, and merged #248, #256, and #260. Exact merge or cumulative main commits passed six-platform package smoke on Ubuntu, Windows, and macOS Intel/Apple Silicon.
Completed three review rounds on #259, finding dual ANN-branch coverage and Windows takeover state-machine gaps. After the author addressed stale failure counts, port exhaustion, and owner identity, exact head af5dbaf passed 9/9 focused tests, typecheck, Prettier, and six-platform package smoke and was approved. Merge order with my overlapping #257 remains explicitly owner-controlled.
Authored fixes #257 and #258 are clean, mergeable, and six-platform green, but remain explicitly pending independent review. This is Collaborator evidence, not a Maintainer or ownership claim.

EvalRepro #31 merged a pinned-revision, hash-only reproduction of the v0.15.0 to v0.16.0 eight-file release contract. The external workflow detected exactly the three intended generated-workflow changes, with its standard matrix and public-source reproduction passing. This records public design-partner validation, not adoption or endorsement.

awesome-ai-security-tools #52 merged the project into its 1k+ Star Watchlist and explicitly classified it as a broader, self-declared readiness gate rather than a security scanner or enforcement control. The entry also records that the project is new, has minimal adoption, and its listed consumer is author-operated. This is Watchlist inclusion and external positioning review, not adoption, certification, or endorsement.