THE OPEN-SOURCE LANDSCAPE

EXAMPLES Example 1: A curated resource list. A community maintains a public list of open-source agent frameworks, tool libraries, and evaluation suites, updated continuously by volunteers. A newcomer opens it and sees the ecosystem grouped by category. Instead of discovering each project by accident, they use it as a map, pick two evaluation tools to try, and note the results in their own notebook. They also check the list's last update date and see that it was updated recently, which increases their trust. Example 2: A reference implementation. A learner clones an open-source minimal agent harness that is only a few small files. They read it from beginning to end in one afternoon and trace one request through the loop, the tool calls, and the stopping rule. When they begin the Part Sixteen capstone, they use it as the seed and change it to fit their own design. Example 3: A community benchmark. A public dataset contains realistic multi-step tasks that many teams run their harnesses against. A team runs the benchmark and learns that its harness does well on short tasks but poorly on long ones. Because everyone uses the same tasks and scoring, the team can compare fairly with unrelated projects and decide what to improve. They also keep a private set of their own tasks to make sure they are not simply tuning for the public exam. Example 4: Risk of relying on open source without review. A popular starter template has not been updated for a year, and its default settings allow the agent to run any shell command with no approval. A team that copies it blindly ships an agent with far more power than it needs. Another team applies the Part Nine review, notices the weak defaults, tightens the permissions, and tests in a sandbox. The second team gets the benefit of the template without inheriting its risk.