autonomous-agents

autonomous-agents is an agent skill about AI systems that decompose a goal, plan actions, execute tools and correct themselves without constant human guidance. It covers agent loops such as ReAct and plan-and-execute, goal decomposition, reflection patterns and production reliability. The skill's central point is that reliability matters more than capability. Each extra decision multiplies the chance of failure, so compounding error rates can defeat an autonomous agent. As an illustration, a 95 per cent success rate per step falls to about 60 per cent by the tenth step. The skill recommends building for reliability first and autonomy second, and it notes that constrained, domain-specific agents with clear boundaries have done better than attempts at autonomy for everything. It treats AI outputs as proposals rather than truth. Use it when you are designing an agent that acts with some independence, and you need to judge where to add checks, boundaries and human review. It is part of the sickn33/agentic-awesome-skills repository on GitHub, which has about 47,600 stars. The listing shows it as free, and it has about 1,200 installs.…