Mods
A plugin and extension mechanism for Claude Code, built on the Function Hooks design pattern. It allows developers to integrate code into and modify the behavior of the software layer surrounding the model (the Harness), including context assembly, tool-call filtering, and interface rendering.
Timeline (last 90 days)
Claude Code officially supports Mods starting from version 2.1.287 and above.
Mods run directly inside Claude Code and can deeply intervene in the agent's behavioral logic.
Mods can intervene just before Claude Code is about to execute an action, modify that action, or even block it entirely.
Mods are loaded once at the start of a session and then run continuously throughout the entire session.
Because Mods run persistently, they can remember state, provide a real-time panel, and pause to ask for user confirmation before executing commands.
Mods can directly intervene in the harness's execution loop itself, inserting custom logic at the critical decision points where the agent determines its actions.
The launch of Mods marks a shift in Anthropic's open strategy for Claude Code—from building an ecosystem around tools to allowing intervention inside the tools themselves.
Components that can intercept and modify agent behavior carry a greater blast radius, and therefore demand more careful design from developers.
The Plugins and Mods mechanisms allow developers to extend the functionality of Claude Code according to their own needs.
Anthropic has launched new features for Claude Code, including Mods, Plugins, Projects, and Tag.
5 more timeline events
All Facts (15)
Because Mods run persistently, they can remember state, provide a real-time panel, and pause to ask for user confirmation before executing commands.
50%UnverifiedMods can directly intervene in the harness's execution loop itself, inserting custom logic at the critical decision points where the agent determines its actions.
50%UnverifiedThe launch of Mods marks a shift in Anthropic's open strategy for Claude Code—from building an ecosystem around tools to allowing intervention inside the tools themselves.
50%UnverifiedComponents that can intercept and modify agent behavior carry a greater blast radius, and therefore demand more careful design from developers.
50%UnverifiedMods run directly inside Claude Code and can deeply intervene in the agent's behavioral logic.
50%UnverifiedMods can intervene just before Claude Code is about to execute an action, modify that action, or even block it entirely.
50%UnverifiedMods are loaded once at the start of a session and then run continuously throughout the entire session.
50%UnverifiedThe Plugins and Mods mechanisms allow developers to extend the functionality of Claude Code according to their own needs.
50%UnverifiedPlugins generally refer to functional modules developed by official or third-party developers and reviewed for quality, focusing on stable integration interfaces; Mods lean more toward community-driven lightweight customization.
50%UnverifiedThe core value of a layered extension architecture lies in the official party maintaining consistency in the core experience while long-tail needs are addressed by the community.
50%UnverifiedAnthropic defines Mods as a plugin that uses Function Hooks, allowing developers to plug in with code and alter the behavior of Harness.
50%UnverifiedThe public repository already contains four built-in Mods: an Agent preloader, a Diff panel, Telemetry, and an organizational security Mod.
50%UnverifiedDevelopers of custom Mods need to first enable the Function Host, and the API may change across different versions.
50%UnverifiedClaude Code officially supports Mods starting from version 2.1.287 and above.
50%UnverifiedAnthropic has launched new features for Claude Code, including Mods, Plugins, Projects, and Tag.
50%