Automate iterative OODA cycles for continuous agent self-improvement.
npx skills add https://github.com/Salahsr01/Cleanbot --skill self-learning-salahsr01
Please help me install this Agent Skill. Skill: self-learning Source: https://github.com/Salahsr01/Cleanbot/tree/main/workspace/skills/self-learning Command: npx skills add https://github.com/Salahsr01/Cleanbot --skill self-learning-salahsr01
Automates the structured self-improvement workflow for an agent, enabling ongoing learning, experimentation, and adaptation to meet objectives.
Inspect current metrics, formulate an improvement hypothesis, run a small experiment in memory/experiments/, and document the results.
High-intent search queries and answers about installing and using this skill.
Automated continuous self-improvement uses an OODA loop of observation, orientation, decision, and action across iterative heartbeat cycles. This allows an AI agent to systematically learn, adapt timing, and expand targets to meet objectives.
A multi-armed bandit approach for agent experimentation systematically tests multiple variants of frameworks or timing adjustments. It dynamically allocates resources to the best performers to optimize objective outcomes.
Structure agent memory by maintaining a centralized playbook and a dedicated directory like memory/experiments/. Document improvement hypotheses, run small tests, and record results to drive future strategies.
Yes, the self-learning workflow explicitly supports trying new frameworks, making timing adjustments, and target expansion. It integrates these actions into the decision and action phases of the OODA cycle.
Start by inspecting current metrics to formulate an improvement hypothesis. Run a small experiment within a memory/experiments/ directory, document the outcomes, and leverage scripts or references to extend capabilities.
StepowskiEricOrganize agent decisions into an OODA loop with ooda-cycle-log.md artifacts.
x402agentRun, monitor, and debug the Seeker daemon and Telegram command flow.
jdrhyneManage a persistent TODO scratch pad via CLI bash script.
GuitarAlchemistSimulate multi-armed bandit algorithms with epsilon-greedy, UCB1, and Thompson sampling.
