Agent 工作流

A practical walkthrough for learning with coding agents while controlling API spending, protecting credentials, and building real engineering skills.

Build a monthly Claude Code budget from measured usage, actual token rates, caching, and the tasks you expect to finish.

Trace a coding agent’s bill from prompts and tool calls to caching, retries, and practical controls for predictable spending.

A practical, seven-step walkthrough for preserving local work, reversing agent damage, and verifying a safe recovery across multiple files.

A practical walkthrough for dividing repository research, implementation, and verification between Gemini CLI and Claude Code.

A practical walkthrough for using coding assistants to produce smaller diffs, clearer tests, and pull requests that reviewers can verify quickly.

A practical walkthrough of Claude Code sessions, CLAUDE.md, automatic memory, and the context you should save before starting fresh.

A seven-step walkthrough for using a coding agent to map a repository, trace behavior, test assumptions, and leave a useful handoff.

A numbered workflow for using Codex on infrastructure-as-code changes while keeping diffs small, plans explicit, and approvals human.

Set up Claude Code in a devcontainer with reproducible tooling, runtime credentials, gateway routing, and sensible permission boundaries.

A practical walkthrough for routing VS Code coding assistants through one API hub without mixing up credentials, models, or protocols.

A practical walkthrough for deciding when a coding agent can run a shell command, needs a safer plan, or must ask for approval.

A numbered walkthrough for instructing coding agents to write deterministic, behavior-focused unit tests that catch meaningful regressions.

A numbered walkthrough for choosing Codex sandbox boundaries, approval policies, network access, and safe automation defaults.

A practical walkthrough for isolating agent edits with Git worktrees, explicit tool paths, sandbox boundaries, and review gates.

A practical walkthrough for choosing Claude Desktop or Claude Code, managing repository access, reviewing changes, and configuring API routing.

A practical walkthrough for isolating two coding agents, assigning ownership, and integrating their changes without repository collisions.

A practical walkthrough for scoping changes, tracing dependencies, testing behavior, and reviewing Claude Code patches in a large monorepo.

A practical numbered walkthrough for using Gemini CLI to handle focused shell automation chores safely.

A numbered walkthrough for reviewing agent-generated changes, from scope and trust boundaries to failure paths, tests, and rollback.

A numbered workflow for planning, testing, deploying, and verifying schema migrations with a coding assistant and explicit human checkpoints.

A practical walkthrough for generating reviewable pull request descriptions with Codex in CI, without granting unnecessary permissions.

A numbered walkthrough for writing concise, testable CLAUDE.md instructions that improve coding-agent decisions without adding noise.

A practical walkthrough for refactoring legacy code with a coding agent, using characterization tests, small diffs, and explicit review gates.

A practical walkthrough of custom-host setup, protocol checks, authentication, model selection, and unchanged local behavior in Gemini CLI.

A practical walkthrough for configuring Codex CLI with a custom provider, checking Responses API support, and testing safely.