Lessons learned • Continuous updating • LegendsThe rules of honesty at this station.
This is not a list of “recommended for all to be installed,” but a map of my real production system. The people who started first.Default Minimum ToolbarRun through a chain; come back here and find an alternative when there are clear limits.
I split it into three floors:ToolsIt’s the software I’m running directly.**Technology.**It’s the bottom of the project running and distributing.Methods and agreementsDecide how the people and Agent collaborate. Combination of the three layers would have a lot of names, but work would not be clearer.
There are only four states:Coreis the default access point for high frequency;Professionalis enabled for a specific task;ExploreThe representative is being tested and cannot pretend to be stable;DiscardThe Representative has withdrawn from the workflow. The tools only belong to me when they enter the stream.
I. Tools I use directly
Each entry is written as “what to do and why to keep; when to switch; public validation; state”. It’s mission map, not software encyclopedia.
Models and dialogue
- ChatGPT / GPT series core: Universal think-through portal, used to launch problems, ICP, cross-references. The long text will look at Claude, and the straight-up scenario will change the model. Authentication:AI is an active source.。
- **Claude.**Second perspective on long text reading, writing discussion and code issues; not switching from model ranking to access when access is unstable. Authentication:AI Reflection and Research。
- DeepSeek.: The Chinese reasoning complements the available scenes in the country; the key facts still go back to their original sources and do not use model answers as evidence. Authentication:AI Use your mind。
- Specializing in bean buns.(b) Daily dialogue and quick attempts in the domestic environment; complex implementation is entrusted to Agent, who carefully studies back to official information. Authentication:AI Reflection and Research。
- Fable Zoo.: try visualizing narratives and content expressions; return to Marktown when value is not clearly shown. Authentication:Thinking and writing。
Agent, editor and model access
- Codex Desktop / CLI core: Multi-file, long, complex execution that runs the test is my default Agent. Just looking at Cursor, hand-held VS Code. Authentication:Why do I make it, Jarvis?。
- Claude Code.: Another strong implementation chain for programming tasks; environment or access is not used when it is unstable. Authentication:Agent execution。
- **WorkBuddy.**Introduction, domestic environment and visualization of office operations; complex warehouse implementation remains a priority, Codex. Authentication:AI, White’s in action.。
- Cursor’s specialty.: Need to keep looking at code, small step changes and cutting models; cross file long assignments to Codex. Authentication:Agent execution。
- **VS Code.**Manual editing, extension of the base of ecological and final takeover; not treating the editor itself as Agent capability. Authentication:Agent execution。
- CCSwitch explores: a candidate tool to manage different model access configurations; no additional layer of configuration is added when the cost of switching is not specified. Open validation: This page records the state of exploration.
- **Diffy explore.**Validation of visual workstreams and Agent product patterns; not yet stable into default production chains. Authentication:Business AI transition methodology。
Writing, knowledge and collaboration
- Typola core: The default entry for Markdown writing, with direct and minimal interference. A large number of double-chained pens were cut, and Obsidian collaborated in the delivery of the cut-off book. Authentication:Thinking and writing。
- Obsidian core: Local knowledge connections and long-term organization; multiple comments, permissions and letters of cut-off upon delivery. Authentication:AI Memory System。
- Flying Book Document / Knowledge Base CoreShare, comment, deliver and collaborate in business knowledge; private, long-term source information is preferred to remain locally. Authentication:Personal knowledge banks are not for yourself.。
- Flying Book Multi-Drive Forms Professional:: Structured records, forms and process collaboration; individual rapid analysis priority Excel/Python. Authentication:Data and office automation。
- Flying Book, CLI.: Automation and reviewability of repeat flyer book operations; one-time small operations directly interface. Authentication:Data and office automation。
- IMA exploration• Validation of the convergence of information and knowledge question and answer experience; failure to create a stable stream of work without acting as a core knowledge base. Open validation: This page records the state of exploration.
Inspection, data and content production
- Playwright core: repeat clicks, cross-views, error and regression checks; change the Command Use when the system primary interface cannot be located. Authentication:Front-end needs for visual reference.。
- Computer Use.: Operation of the system interface and visual review as a human being; no one can use an accident as a stabilization test. Authentication:Check and release。
- Chome DevTools / CurlCheck Dom, Network, Console and HTTP smoke; they give evidence and do not replace real user processes. Validation: This station is open for acceptance at the front end.
- Excel / Python coreExcel is responsible for visible, spotable analysis, Python is responsible for cleaning and batch processing; and slicing books when multi-person collaboration is required. Authentication:Data and office automation。
- WPS / Word / PDF: Office delivery format for realistic recipients; self-writing remains a priority for Marktown. Authentication:Data and office automation。
- Clip-and-show profession: Video clips requiring manual judgement and rapid adjustment; Remotion is considered when replicable, code generated. Authentication:Content production。
- Remotion: Reusable, batchable video generation; one-off video is not highly technical. Authentication:Content production。
- AI review + release feedback core: Repetition, counter-example and factual risk, then correction of content with real feedback; not allow AI to sharpen individual judgement into standard answers. Authentication:AI, write the truth.。
Personal system
- Jarvis core.• Maintain continuity, attention and project context, manage multiple Agents; long-term memory must be confirmed without making life decisions for me. Authentication:Why do I make it, Jarvis?。
- StarPeer Professional: High-value open-source contribution opportunities identified and followed up; not used for mass brushing of low quality PR. Authentication:Codes and collaboration。
- Tomato. Todo core.: Constraint attention to current tasks; too small a task without an additional record-keeping burden. Authentication:My methods of work。
II. Project technology warehouse
These are the floors under which the project is run and distributed, not the software that is recommended to be fully installed by ordinary readers.
- Git / GitHub / GitHub CLI: acceptance of versions, collaboration, automation and public delivery. The local small experiment also retains Git, with open collaboration on GitHub. Validation: This station and open warehouse.
- Next.js / Vite / VitePress / Quartz: Select the Web and the document by product form to construct the base, without no justification for ” technology update ” . Authentication: You’re looking at Jim Wiki using Quartz.
- Prisma / PostgreSQL: application floors requiring relational data, migration and reliable queries; static content is not included in the database. Validation: Open project practice.
- Docker / NginxEnvironmental sealing, service deployment and reverse agents; static Pages do not add server complexity when they can be resolved. Authentication:Check and release。
- GitHub Pages / CodexitesThe former issues current static stations, while the latter are used for the fast-track site prototype, and the full deployment is replaced when there are back-end and authority requirements. Could not close temporary folder: %s
- Log, screenshot, backup and rollback core: enable location, proof and return after publication; no production release omits the chain of evidence. Authentication:The rules of true failure.。
Methodology and agreement
These are not software names, but they affect delivery quality more than a change in software.
- Target success criteria Investigation plan Implementation testing feedback deposition: Define the end condition and avoid Agent replacing the result with a nice report.
- Master, project manager.: I’m only going to control the target, and I’m going to hold the project boundary and report back; I’m not going to let a group of Agent make a scene in a group chat.
- Maker + Checker: The achiever cannot use his/her own automation as a substitute for independent acceptance, and the key vision is to see the true screenshot.
- Worktree / Step / Rollback Point: Parallel tasks are segregated, changes are submitted to range, and when failed, the last available status is returned.
- GitHub Helipad collaboration: Task packages, diff, tests, screenshots, reports and responses are all searchable products and do not depend on the oral context in the chat window.
My final judgment.
Full-scale does not amount to a stack of names. My default mill is responsible for getting people to start, and the complete map is responsible for explaining how I switch when I encounter a different task; the technology vault is responsible for running the product, and the methods and protocols are responsible for making it credible.
If you haven’t run through a full production chain, don’t go through the page at once. I’ll do what I want, and then I’ll do what I want; when you’re really constrained, the alternative means will make sense.