Nội dung
- Key points
- This repo is solving a very real AI agent pain point
- Why the AI community is paying close attention to the “Manus-style” approach
- What makes the repo’s growth numbers create a FOMO effect
- A perspective for the Vietnamese market
- Conclusion: a small repo that taps into a big trend
- References
- Frequently asked questions
OthmanAdi’s planning-with-files is an open-source repo that helps AI agents plan using markdown files in a crash-proof way, so plans survive context loss or the /clear command. The repo has about 24,185 stars and supports more than 60 agents via the SKILL.md standard, helping agents work for longer without breaking momentum.
Key points
- Planning with markdown files helps agents retain state and survive context loss or the /clear command
- It has a deterministic completion gate: a rule-based completion check that reduces cases where an agent thinks it is done but is not
- Manus-style: multiple agents coordinate and share state on disk
- Compatible with Claude Code, Codex CLI, Cursor, Kiro, OpenCode and more than 60 agents via the SKILL.md standard
- Reached ~24,185 stars, ~2,096 forks, and ~4,031 stars per month despite being only about 6 months old
For marketers and growth teams experimenting with AI for content creation, automation, and campaign operations, a repo that “remembers” can make a big difference. planning-with-files by OthmanAdi is drawing attention as it has reached 24,185 stars, 2,096 forks, and a growth rate of about 4,031 stars per month. The repo is only about 6 months old and was last updated 14 days ago. These numbers show that the community is strongly interested in a more durable way for AI agents to work, especially when long-running tasks are easily interrupted by context loss.
This repo is solving a very real AI agent pain point
The standout feature of planning-with-files is its file-based planning approach, designed to be “crash-proof.” Instead of relying entirely on the model’s short-term memory, the plan is written in markdown so it can survive context loss or when the user runs /clear. For long agentic tasks, this is a major advantage because AI does not just need to answer quickly; it also needs to reach the finish line in a controlled way.
The repo also emphasizes a deterministic completion gate — simply put, a rule-based completion check that reduces the chance of an agent “thinking it is done” when it has not actually met the requirements. In real work environments, especially product work, marketing automation, or coding assistants, that consistency is just as valuable as speed.
Why the AI community is paying close attention to the “Manus-style” approach
According to the source description, this repo follows a Manus-style approach: multiple agents work together, with shared state stored on disk rather than each run being a separate process. This model fits the current agent trend: not just “chatting with AI,” but organizing AI as a work system with planning, state, and task handoffs.
The repo’s appeal also lies in its practicality. It is not just an idea; it is designed to work with popular tools such as Claude Code, Codex CLI, Cursor, Kiro, OpenCode and more than 60 agents through the SKILL.md standard. In other words, users are not locked into a single ecosystem, which is especially important for technical teams and marketing teams that want to experiment quickly.
What makes the repo’s growth numbers create a FOMO effect
In the open-source world, attention often directly reflects usefulness, or at least timeliness. With 24,185 stars and an average of about 4,031 stars per month since creation, this repo is showing strong momentum relative to its age of only about 6 months. In addition, 2,096 forks suggest that many people want to deploy it themselves, modify it, or integrate it into their own workflows.
Being updated 14 days ago is also an important signal: this is not a “viral and abandoned” project, but one that is still being maintained relatively actively. For teams looking for a reliable AI agent solution, this is the kind of repo that often makes users want to try it immediately so they do not miss a new working standard.
A perspective for the Vietnamese market
For Vietnamese businesses, especially marketing, growth, and product teams using AI to speed up content production, market research, or operations support, the big question is not “Is AI smart?” but “Can AI keep the workflow going?” Long workflows such as building landing pages, creating content sequences, analyzing competitors, or coordinating multiple automated tasks all require the ability to save state and resume from the right point after an interruption.
That is why file-based planning models like planning-with-files are worth following. Not every team needs to adopt them immediately, but the idea of an “agent with a plan, checkpoints, and durable state” will be useful for Vietnamese businesses that want to move from using AI as a chat tool to using AI as a real operational component. It is also a good cue for agencies and in-house teams that want to standardize their agentic AI experimentation process before scaling up.
Conclusion: a small repo that taps into a big trend
planning-with-files stands out not only because of its high star count, but because it addresses a very real need in the AI agent era: long-running work, no context loss, a clear completion mechanism, and support for a multi-tool environment. As many teams look for ways to bring agents into real workflows, this is a repo that Vietnamese tech marketers, product managers, and developers should watch closely.
If you are experimenting with AI for content, automation, or operations, the most important idea from this repo is: if you want an agent to work seriously, give it durable memory and a clear control process.
See more marketing news and guides at https://marketing365.vn.
Follow more updates from Marketing365 to stay up to date with the latest marketing trends.
Read more articles in the Digital Trends category.
References
Frequently asked questions
What is planning-with-files used for?
The repo helps AI agents save work plans as markdown files instead of relying only on the model’s short-term memory. This allows agents to continue the workflow correctly after losing context, making it suitable for long tasks such as content creation, automation, or competitor analysis.
Which AI tools can planning-with-files work with?
The repo supports popular tools such as Claude Code, Codex CLI, Cursor, Kiro, OpenCode and more than 60 other agents through the SKILL.md standard. Users are not locked into a single ecosystem.
Why is this repo attracting the AI community?
It addresses a real need in the AI agent era: long-running work without context loss and a clear completion mechanism. With about 24,185 stars, 2,096 forks, and a recent update, it shows strong momentum despite being only about 6 months old.
What should Vietnamese businesses pay attention to in this repo?
The big question for marketing and product teams is not whether AI is smart, but whether AI can keep the workflow going. The idea of an agent with a plan, checkpoints, and durable state helps Vietnamese businesses move from using AI for conversation to using AI as a real operational component.
