B4.run: capabilities, use cases, and sources

Build TypeScript agents with tools, workspaces, sandbox execution, and approval. Inspect a real recorded repair and run the blueprint yourself. Key features: - /research (agent) - /research/subagents/researcher (agent) bash npm test Runs both packages' suites, --dry-run creates no model and performs no write;, --record captures real-model responses as committed fixture files that plain b4 eval replays, , where ctx is B4ToolContext from @b4run/sdk — carrying signal, threadId?, middleware?, and fs: src/app/(public)/hello/[tenant]/tools/greet, ", "suggestedPattern": "["/support","researcher"]" } } The input preview is bounded, display-only text, ", }) Tools live next to the agent — index, ", tools: { approve: ["exportForReview"] }, systemPrompt · instructions folded}) Copy source 02Workspaces + sandboxes Give it somewhere to work, ) A minimal tool src/app/(public)/hello/[tenant]/tools/greet.