MicrocourseIntermediateBranches and Merges in GitHubYou will understand what branches and merges are for, decide when a branch is worth making, and direct an agent to bring work on stray branches back into main without losing anything.25 min12 insightsMembers
MicrocourseIntermediateSingle Source of Truth for AI ToolsYou will put one client-email rule in one official file, connect Claude and Codex to it, test both tools on the same email, and catch old copies without deleting genuine project exceptions.16 min13 insightsMembers
MicrocourseIntermediateConnect a Service Without a Built-In ConnectorYou will confirm that the service publishes an application programming interface (API), that the API covers your job, and that your runtime, the app or environment sending the request, supports one of the service's authentication methods. When those checks pass, you will create a narrowly scoped credential, keep it outside chats and source files, test one small read, and revoke or replace it when it is exposed or no longer needed.18 min10 insightsMembers
MicrocourseIntermediateSplit a Multi-Stage Job Across AgentsYou will decide whether one agent or several fit the job, map dependencies before work starts, run only independent parts in parallel, give each helper one narrow job and the evidence it needs, combine the pieces, and validate the finished result. You will treat each arrangement as a proposal to test rather than a universal workflow.18 min10 insightsMembers
MicrocourseIntermediateGenerate, Compare, and Combine Better DraftsYou will generate candidates that use deliberately different strategies, compare them against written criteria and representative tests, and select the complete candidate that performs best on those tests. You will borrow a part from another draft only when the combined version passes the same checks, and you will stop when another round no longer improves the measured result.15 min9 insightsMembers
MicrocourseIntermediateCheck AI Work Against Written StandardsYou will set up an independent verifier: a separate check of finished work that starts with direct tests, calculations, and source checks. For qualities that still require judgment, you will calibrate a reviewer on known failures and clean cases, track misses and false alarms, require evidence for every finding, and stop or escalate according to the consequence of an error.16 min9 insightsMembers