Record user corrections and preferences into long-term rule files.
npx skills add https://github.com/picsky/flowos --skill self-improving-picsky
Please help me install this Agent Skill. Skill: self-improving Source: https://github.com/picsky/flowos/tree/main/templates/skills/self-improving Command: npx skills add https://github.com/picsky/flowos --skill self-improving-picsky
This Skill captures user corrections and preferences, confirms ambiguous intent, and converts repeated feedback into durable long-term rules to reduce future friction.
Tell the agent when something is wrong or your preference (for example: "不对，我不喜欢太长的回复，应该简短一点，以后都这样。").
High-intent search queries and answers about installing and using this skill.
To make an agent learn from corrections, you provide direct feedback during conversations. The agent detects preference signals, confirms ambiguous intent, and converts repeated feedback into durable long-term rules to reduce future friction.
Feedback learning works by tracking how often the same correction occurs and promoting it to long-term principles after a repetition threshold is reached. It records structured understanding of required changes to improve future agent behavior.
The best way to capture preferences automatically is through silent rhythm signal collection. The agent extracts schedule and energy-related info into rhythm memory without interrupting your flow, while detecting explicit preference signals to update behavior.
You can apply corrections universally or to a specific agent. When the scope of your feedback is unclear, the agent asks targeted questions to pin down whether the preference applies to the current task, a specific agent, or universally.
Agent behavior rule mining does not require manual memory updates. It relies on deterministic memory reads and writes to automatically update corrections and related long-term rule files using repetition-based promotion.
picskyIntroduce the “如流” assistant and collect identity, sleep rhythm, and time blocks.
build000rTransform scattered agent sessions into persistent cross-agent procedural memory.
rakibulismInitialize a productivity workspace by creating TASKS.md, CLAUDE.md, memory files, and dashboard.html.
tiankong0101-byteTrack interaction outcomes and refine skill behavior from learned patterns.
