Disabling Features

Weave lets you selectively disable agents, tools, governance hooks, and skills. All disabling is done via top-level disabled_* arrays in your config — no need to delete or modify anything else.

Disable Arrays

{
  "disabled_agents": ["warp"],
  "disabled_tools": ["webfetch"],
  "disabled_hooks": ["context-window-monitor"],
  "disabled_skills": ["some-skill"]
}

Each array takes the names (string identifiers) of the things to disable. Entries from your user config and project config are unioned — so if your user config disables context-window-monitor and your project config disables webfetch, both are disabled.

Disabling Agents

Remove agents you don't need from the system entirely:

{
  "disabled_agents": ["warp", "spindle"]
}

Alternatively, you can disable an agent via its per-agent config:

{
  "agents": {
    "warp": { "disable": true }
  }
}

Both approaches have the same effect.

Disabling Tools

Remove specific tools from all agents:

{
  "disabled_tools": ["webfetch", "bash"]
}

WARNING

Disabling tools affects all agents, not just one. If you want to restrict tools for a specific agent only, use agents.<name>.tools with per-tool toggles instead.

Disabling Hooks

Weave's governance hooks run automatically during agent execution. You can disable individual hooks if they interfere with your workflow:

{
  "disabled_hooks": ["context-window-monitor", "verification-reminder"]
}

Available Hooks

Hook	Purpose
start-work	Parses plan files and switches to Tapestry when /start-work is called
work-continuation	Handles work resumption signals: idle work nudges and post-compaction work recovery
workflow	Handles workflow start, workflow commands, and workflow continuation behavior
todo-continuation-enforcer	Finalizes leftover in_progress todos automatically, or prompts as a fallback if enabled
compaction-todo-preserver	Preserves and restores todo state across compaction
pattern-md-only	Blocks Pattern from writing non-.md files (plan discipline enforcement)
context-window-monitor	Warns when context window usage is high
write-guard	Validates write targets before file operations
rules-injector	Loads AGENTS.md and .rules files into agent context
keyword-detector	Detects specific keywords in messages and triggers actions
verification-reminder	Prompts agents to run verification steps before completing
analytics	Tracks session activity (tool usage, delegations) and generates project fingerprints

INFO

Hook names use kebab-case in configuration. For example, use context-window-monitor in your disabled_hooks array, not contextWindowMonitor.

Continuation-related hooks

Preview

The newer continuation config surface and its interaction with recovery/idle behavior are currently in preview.

Continuation behavior is split across several controls.

• Disable work-continuation to turn off both idle work nudges and post-compaction work recovery
• Disable workflow to turn off workflow continuation behavior entirely
• Disable todo-continuation-enforcer to stop automatic todo cleanup
• Disable compaction-todo-preserver to stop preserving todo state across compaction

If you want to tune continuation behavior without disabling a feature entirely, use the continuation config block instead of disabled_hooks.

{
  "continuation": {
    "recovery": {
      "compaction": true
    },
    "idle": {
      "enabled": false,
      "work": false,
      "workflow": false,
      "todo_prompt": false
    }
  }
}

See Execution & Continuation for the full behavior model.

Common Disables

context-window-monitor is the most commonly disabled hook — it produces informational noise during long sessions. verification-reminder is useful to disable when you're doing rapid exploratory work and don't want the extra checkpoint prompts. analytics can be disabled if you don't want session tracking or project fingerprinting (see Analytics).

Disabling Skills

Exclude specific skills from being loaded, even if they exist in a skill path:

{
  "disabled_skills": ["tdd", "some-inherited-skill"]
}

This is useful when you've inherited skills from a shared skill library (e.g., a monorepo) and want to opt out of specific ones for a particular project.