Notas de sprint
Tutoriais rápidos sobre agentes de programação, escolha de modelos e controle dos gastos com tokens.
为文档数据提取选模型：从文件检查到上线的六步先区分文本与扫描文档，再用统一字段、真实样本和成本记录筛选模型，建立可追溯、可校验的文档数据提取流程。
七步比较模型编写单元测试的实际能力通过统一任务、约束提示、编译检查、缺陷验证和维护性审查，逐步比较模型生成单元测试的质量，找到适合项目的使用方式。
客服机器人如何选模型：从常见问答到复杂投诉的七步实战用七个步骤比较轻量模型与推理模型在客服聊天中的分工，建立测试集、知识检索、升级路由和成本核算，避免只看单次调用价格。
哪个模型最擅长处理泛型：六步选出可靠助手通过统一题目、检查类型推断、验证条件类型和运行正反例测试，挑选真正适合处理泛型代码的模型，而不是只看回答是否流畅。
混元与小米模型接入中文应用：六步实战围绕中文客服与知识库问答，分步完成模型核对、样本设计、统一接入、提示词约束、双模型评测与上线保护，让选型依据来自真实业务。
多模态调试笔记：从截图证据到回归验收的七步实操围绕截图、录屏、日志和代码，逐步组织多模态调试材料，区分事实与假设，用最小实验验证根因，并留下可复现、可交接的调试笔记。
双模型架构讨论：七步对照方案与决策依据用同一份需求逐步对照两款模型的架构建议，通过独立提案、交叉质询、故障推演与验证任务，把讨论沉淀为可执行的架构决策。
上下文窗口多大才够用：从代码修复到文档问答的六步实操通过代码修复与文档问答两个项目场景，逐步理解上下文窗口、词元预算、输出预留与资料筛选，判断什么时候需要更大窗口，什么时候应该先精简输入。
七步比较推理模型排查并发缺陷的能力从丢失更新到死锁，用统一任务、受控交错、最小补丁和回归测试，逐步比较推理模型处理并发缺陷的能力，而不是比较答案是否听起来专业。
Which Model to Pick for SQL Query TuningChoose a model for SQL tuning with a repeatable workflow: prepare query context, compare diagnoses, test rewrites, and track token costs.
六步选出适合代码审查评论的模型从审查目标、上下文边界和真实缺陷出发，用小样本对照、误报检查与分级路由，选择能给出准确、可执行代码审查评论的模型。
Claude Fable 5.1: Where Premium Pays OffA step-by-step guide to testing Claude Fable 5.1 on difficult development work and deciding when its premium cost is justified.
MiniMax M3 for Multilingual Product Copy: 7 StepsBuild a repeatable MiniMax M3 workflow for localized product copy, with clear briefs, protected placeholders, automated checks, and native review.
Grok 4.7 Conversational Prototypes: 7 StepsBuild a Grok 4.7 conversational prototype with SwiftRelay, clear prompts, explicit conversation history, safe mock actions, and repeatable tests.
GLM-5.3 for Chinese-English Projects: 7 StepsBuild a practical GLM-5.3 workflow for Chinese-English projects, from terminology and prompts to localization checks and release reviews.
DeepSeek V4 Pro for Algorithms: A Step-by-Step GuideUse DeepSeek V4 Pro for algorithm-heavy tasks with precise prompts, correctness checks, differential testing, and a practical SwiftRelay setup.
Read Large Codebases with Gemini 3.1 ProA step-by-step workflow for mapping repositories, tracing execution paths, and checking Gemini 3.1 Pro's findings against real code.
Pair Programming a Python Pipeline with Claude CodeBuild a testable Python data pipeline with Claude Code through scoped prompts, explicit data contracts, fixture tests, and careful review.
Claude Code on Windows: PowerShell Setup NotesSet up Claude Code in PowerShell, connect SwiftRelay, preserve environment settings, and troubleshoot Windows paths and authentication.
Start a Weekend Side Project with CodexA step-by-step weekend plan for building with Codex, from a focused project brief to tested features and a deployable demo.
Switch Coding Agent Models Without Losing ContextA step-by-step guide to switching coding agent models mid-project while preserving context, protecting changes, and checking results.
Draft API Reference Pages with Claude CodeA step-by-step workflow for turning source code into API reference drafts, with evidence checks, tested examples, and focused reviews.
A Coding Agent Routine for Flaky E2E TestsA step-by-step routine for coding agents to reproduce flaky end-to-end tests, isolate causes, make minimal fixes, and verify results.
Gemini CLI: Shell One-Liners and Bulk RenamesUse Gemini CLI to draft shell one-liners, preview bulk file renames, check collisions, and apply changes with a reviewable workflow.
Coding Assistant Prompts for Legacy JavaScriptRefactor legacy JavaScript with focused coding assistant prompts, characterization tests, small patches, and explicit behavior checks.
Codex Commit Messages Reviewers Actually ReadA step-by-step Codex workflow for drafting commit messages that explain intent, stay grounded in the diff, and help reviewers focus.
When to Restart a Claude Code SessionA step-by-step guide to spotting session drift, preserving useful work, and restarting Claude Code with a focused, testable brief.
Coding Agent Etiquette for Shared RepositoriesA practical walkthrough for keeping coding agents scoped, reviewable, and respectful of teammates in a shared repository.
Run Two Coding Agents in Separate Git WorktreesA step-by-step workflow for isolating two coding agents, assigning bounded tasks, reviewing changes, and merging safely with Git worktrees.
Review a Coding Agent Diff Before You Accept ItA step-by-step workflow for checking scope, correctness, security, and tests before accepting changes from a coding agent.
Claude Code Slash Commands That Cut Repeated TypingTurn repeated Claude Code prompts into reusable slash commands with arguments, focused checklists, and shared team workflows.
Let a Coding Agent Write Safer Database MigrationsA step-by-step workflow for agent-written database migrations, with scoped permissions, staged changes, realistic tests, and human approval.
Keep Your Coding Assistant Focused, One Task at a TimeA step-by-step workflow for scoping coding tasks, setting boundaries, checking changes, and keeping assistant sessions on track.
Pair With a Coding Agent on an Unreproducible BugTurn an intermittent bug report into testable hypotheses, targeted instrumentation, and a verified fix with a coding agent.
Teach Codex Your Repository with AGENTS.mdA practical walkthrough for mapping your repository, documenting commands, and verifying that Codex follows your AGENTS.md instructions.
Codex CLI: From Install to First Merged ChangeInstall Codex CLI, configure access, make a focused fix, verify the diff, and merge your first tested pull request.
Set ANTHROPIC_BASE_URL Without Breaking Your ShellConfigure Claude Code with SwiftRelay using a private environment file, safe shell-profile edits, syntax checks, and a clear rollback.
Claude Desktop Custom Endpoint: Check These FirstA step-by-step guide to checking Claude Desktop gateway settings, API compatibility, credentials, model names, and request routing.
Your First Claude Code Session on a New MachineInstall Claude Code, configure SwiftRelay, verify your connection, and make a small, tested change on a brand-new machine.