cursor/cookbook

Features

• Coding Agent Orchestrators - Runs coding agents from TypeScript code or a terminal, streams output live, and sequences dependent tasks so independent work runs in parallel.
• Coding Agents - Starts coding agents from apps, scripts, and workflows through a TypeScript API that works against local workspaces and cloud runtimes.
• Agent Progress Boards - Lists cloud coding agents as kanban cards grouped by status or repository, previews artifacts, and launches new agents from a repository and prompt.
• Native Speech-to-Speech Agents - Connects caller audio directly to a single speech-to-speech model that listens, reasons, calls tools, and speaks replies without separate transcription and synthesis stages.
• Realtime Speech-to-Speech Agents - Uses a single realtime speech-to-speech model that takes caller audio and speaks replies directly, calling typed tools mid-conversation without separate transcription or synthesis stages.
• Customer-Managed Cloud Agent Workers - Runs agent workers in customer-managed environments to clone repositories, run builds and tests, and reach internal services while orchestration stays hosted.
• Agent Event Hooks - Runs configured hook scripts on agent events such as prompt submission, shell commands, and file edits to log, block, or follow up, failing closed on errors.
• Agent Task Dependency Resolvers - Executes a graph of dependent agent tasks ordered by dependency so each step runs only after its prerequisites finish.
• Agent Worker Deployments on AWS - Provides setup guides for running agent workers on AWS virtual machines with Docker, serverless containers, or Kubernetes, including trade-offs and troubleshooting.
• Worker Connections - Connects self-hosted cloud agent workers outbound over HTTPS to the hosted orchestration service, so no inbound access to the worker host is needed.
• Self-Hosted Agent Servers - Guides running cloud agent workers on customer-managed AWS servers using containers on virtual machines or serverless containers.
• Dependency-Ordered Task Execution - Sorts subtasks into dependency ranks, runs each rank in parallel, and injects parent output snippets into child prompts so dependents wait for prerequisites.
• Parallel Subagent Orchestrators - Splits a task into a dependency graph and runs the pieces in parallel across local subagents with live status.
• Dependency-Aware Task Orchestration - Sorts subtasks into dependency ranks and runs each rank concurrently so independent work fans out and dependents wait.
• Agent Run Event Streams - Streams events as an agent run progresses while leaving prompts, models, cancellation, artifacts, and conversation state under the caller's control.
• Agent Governance - Ships hook-based examples that audit agent prompts, shell commands and file edits, block sensitive prompts, restrict models by repository, and keep skills aligned with code.
• CLI Coding Agent Hosts - Runs a coding agent from a command-line tool so developers can hand tasks to an agent from the terminal.
• Cloud Agent Launchers - Lists cloud coding agents as kanban cards by status or repository, previews artifacts, and launches new agents from a repository and prompt.
• Cloud Agent Project Prototyping - Spins up agents in a sandboxed cloud environment to scaffold new projects and refine ideas from a web interface.
• Cross-Language Agent Bridges - Lets programs in other languages run a single agent turn by starting a bridge process and exchanging JSON messages.
• Headless Agent Runs - Creates an agent in a script, sends a fixed prompt, and waits for the run to finish without an editor.
• App Building Agents - Builds new apps from a plain-language description by running a coding agent inside a chat application that writes the code and refines it step by step.
• Parallel Task Graph Execution - Splits work into a dependency graph, runs subagents in parallel by rank, passes upstream results to dependents, and shows live progress on a canvas.
• Terminal Coding Agents - Demonstrates a terminal agent loop calling a hosted model with file and shell tools confined to one workspace folder.
• Appointment Scheduling Systems - Finds open slots, registers patients, and books, lists, cancels, or reschedules appointments during a live voice conversation.
• Hot Reloading Previews - Re-renders a preview frame live as agents write streamed output into a canvas file that recompiles on every write.
• Self-Hosted Infrastructure - Guides running cloud agent workers on customer-managed AWS servers or containers with autoscaling and outbound-only HTTPS connections to the hosted service.
• Model Usage Blocklists - Blocks prompt submissions when both the chosen model and the git origin repository name match configured substring blocklists, failing closed on errors.
• Prompt Secret Scanning - Scans submitted prompts for secrets or sensitive data and blocks them before reaching the model, optionally deferring to a data-loss or secret-scanning service.
• Agent Read and Write Folder Confinement - Confines agent file and shell tools to one workspace folder, rejecting absolute paths, parent traversal, and symlink escapes while capping output, run time, and tool rounds.
• Agent Tool Auditing - Records each agent action through project hooks into an audit log so teams can review what the agent did and when.
• Workspace Path Sandboxes - Resolves every file and shell tool path inside one workspace folder, blocking absolute paths, traversal, and symlink escapes while capping output, run time, and tool rounds.
• Agent Activity Audit Trails - Records prompt submissions, shell commands and results, and file edits as structured entries to a local file or an internal audit service.
• Agent Hook Audit Logs - Writes structured line-by-line records of prompts, shell commands, shell results, and file edits, and can forward them to internal logging or monitoring services.
• Line-Delimited JSON Protocols - Exchanges newline-delimited JSON messages with a child bridge process so any language with standard JSON support can drive a coding agent turn.
• Agent Interaction Streams - Streams an agent's responses and tool usage live into a chat window so users can follow the work as it happens.
• Task Progress Streams - Streams agent output token by token into a canvas file so each task card updates live as work progresses.

Star history