yuanrong-datasystem

Use when reviewing yuanrong-datasystem code, tests, scripts, docs, diffs, commits, pull requests, PR descriptions, or design changes for correctness, design-contract compliance, internal/public API design quality, naming clarity, developer experience, abstraction boundaries, module locatability, production diagnosability, misuse prevention, ownership/lifecycle safety, hot-path performance, concurrency and C++ safety, persistence, recovery, public API/config/docs coverage, Bazel/CMake build support, sensitive-information exposure, testing, operability, risk-calibrated comment quality, context-update risk, and GitCode PR review-comment publishing.

Analyze DataSystem trace bundles for slow latency or errors. Use for gzip/tar trace packages, worker/client trace IDs, RPC deadline exceeded, rpc slow, client summary, RemotePull, Get/Set/Create/Publish, URMA_ELAPSED_*, worker/time/flow/breakdown aggregation, and CodeGraph-backed source mapping.

Create GitCode Pull Requests for openeuler/yuanrong-datasystem or other GitCode repositories by calling the GitCode OpenAPI. Use when the user asks to create, submit, or open a PR for a pushed branch, especially after preparing docs or code commits that must target doc_pages, master, or another GitCode branch. For openeuler/yuanrong-datasystem, the PR body must follow .gitee/PULL_REQUEST_TEMPLATE/PULL_REQUEST_TEMPLATE.zh-cn.md.

Use when a gitcode.com PR's /retest finished and you need to know which checks failed and why. Given a PR number, fetches openeuler-ci-bot comments, parses the check result table, identifies FAILED checks, extracts console URL, fetches console output, and pinpoints failing test cases + actual error messages. Saves 10+ manual API/web fetches per diagnosis.

Analyze KVCache logs and generate interactive HTML reports with ECharts charts. Two scripts: access log analysis (QPS, latency, errors) and worker resource log analysis (memory, thread pools, cache hit rate, ETCD). Triggers: log analysis, access log, resource log, KVCache report, error analysis, QPS trend, latency P99, hit rate, thread pool, memory usage, worker resource, performance analysis, log report generation.

Build and refresh the openYuanrong datasystem online Chinese documentation. Use when the user asks to update, refresh, or publish the online Chinese docs, especially when Codex needs to compile docs/source_zh_cn from the latest upstream master, sync build_zh_cn/html into doc_pages/docs/zh-cn/latest, create a documentation refresh commit, push a refresh branch to a fork, and automatically open a GitCode PR using $ds-create-pr.

Fetch GitCode issues for yuanrong-datasystem, redact sensitive details, and produce a structured task spec for AI-assisted development. Use when the user provides an issue number or URL and asks to analyze, triage, start work from, or prepare an implementation plan for that issue.

Fetch, inspect, reply to, resolve, and verify GitCode Pull Request review comments for yuanrong-datasystem or compatible GitCode repositories. Use when the user asks to pull PR comments, address review comments, reply under each review discussion, mark comments resolved, verify resolved state, or continue after PR review feedback.

Run configured validation for yuanrong-datasystem changes, including local command planning, remote validation through user-provided SSH targets, build.sh/CMake/Bazel commands, and targeted CTest runs. Use when Codex needs to validate code changes, prepare PR verification evidence, or choose tests after implementation. Requires private server details to be supplied through a local config file, not committed to the repository.

Use when implementing, debugging, refactoring, designing, or answering codebase questions that touch yuanrong-datasystem runtime, client, worker, master, common infra, public/internal APIs, developer-facing abstractions, module boundaries, production diagnosability, hot-path performance, concurrency, memory safety, persistence, recovery, build/test behavior, or repo-context updates.

Use when Codex is about to claim yuanrong-datasystem work is complete, tests pass, context is updated, a commit is ready, or a PR is ready after changing files or reviewing code.

Use this skill when diagnosing RDMA/UCX performance, latency, throughput, submit/flush bottlenecks, UCP worker scaling, batch get/pull remote latency, or UCX resource lifetime crashes in the datasystem codebase.