PinSay
A page-comment automation tool designed for AI Agents. It operates through three components—Skills, CLI, and MCP—enabling AI Agents to translate page design comments directly into code changes. It includes built-in security boundaries to prevent prompt injection attacks.
Timeline (last 90 days)
PinSay is a set of MCP (Model Context Protocol) tools for AI Agents, with its core positioning as enabling AI programming assistants to directly participate in code review workflows
PinSay's slogan is: PIN it, Say it, Your agent ships it
PinSay's configuration should be added to user-level MCP settings rather than project-level, to avoid being committed to the codebase
PinSay's toolset is divided into two groups: read operations and write operations
PinSay's read operations include listing comments, retrieving the pending queue, and fetching the source file content corresponding to resolved items
PinSay's write operations will never execute a push; final publishing control remains in human hands
PinSay treats the text content of comments as untrusted data to reduce the risk of prompt injection attacks
PinSay stipulates that errors are returned in the form of error codes rather than natural language descriptions
PinSay's commit style can be configured at the project level, with the option to merge into a single commit or generate a separate commit for each comment.
In PinSay's page component, the status of a processed comment changes to applied, and its auto-reply displays the tools and models used.
10 more timeline events
All Facts (20)
PinSay is a set of MCP (Model Context Protocol) tools for AI Agents, with its core positioning as enabling AI programming assistants to directly participate in code review workflows
50%UnverifiedPinSay's slogan is: PIN it, Say it, Your agent ships it
50%UnverifiedPinSay's configuration should be added to user-level MCP settings rather than project-level, to avoid being committed to the codebase
50%UnverifiedPinSay's toolset is divided into two groups: read operations and write operations
50%UnverifiedPinSay's read operations include listing comments, retrieving the pending queue, and fetching the source file content corresponding to resolved items
50%UnverifiedPinSay's write operations will never execute a push; final publishing control remains in human hands
50%UnverifiedPinSay treats the text content of comments as untrusted data to reduce the risk of prompt injection attacks
50%UnverifiedPinSay stipulates that errors are returned in the form of error codes rather than natural language descriptions
50%UnverifiedPinSay's commit style can be configured at the project level, with the option to merge into a single commit or generate a separate commit for each comment.
50%UnverifiedIn PinSay's page component, the status of a processed comment changes to applied, and its auto-reply displays the tools and models used.
50%UnverifiedPinSay allows AI Agents to directly convert comments marked on a page into code changes.
50%UnverifiedIn the PinSay workflow, it is the developer who reviews the changes and executes the push; the Agent is only responsible for generation and staging.
50%UnverifiedPinSay provides an MCP server, allowing the Agent to perform the same operations via tool calls rather than the command line.
50%UnverifiedPinSay operates through the coordinated work of three components: Skills, CLI, and MCP.
50%UnverifiedWhen the init command is executed, PinSay's skill is installed to the directory location where the Agent searches for tools.
50%UnverifiedAfter running the apply pinsay comments command, the Agent first outputs an execution plan explaining which files each comment will touch, and then proceeds with execution.
50%UnverifiedThe Agent reads the prompt, the team's AI rules, and design tokens from the CLI, ensuring generated code adheres to the team's existing conventions.
50%UnverifiedPinSay treats every comment embedded in the page as untrusted data to defend against prompt injection.
50%UnverifiedPinSay delegates the actual code writing task to a cheaper model, while the planning and review stages are handled by a more capable model.
50%UnverifiedAfter each comment is processed, the mark command executes a commit and records the tool, model, and role that completed the work.
50%