How to Use the i-have-adhd Pre-Send Check Feature in Cursor

The i-have-adhd pre-send check automatically intercepts AI-generated responses before they reach the chat, enforcing length limits and rewriting content that contains distraction keywords to ensure ADHD-friendly communication.

The i-have-adhd skill is a Cursor-compatible extension designed to help users maintain focus during AI-assisted workflows. According to the ayghri/i-have-adhd repository, the pre-send check serves as a gatekeeper that evaluates every draft response against neurodiversity-friendly style rules before delivery.

What the Pre-Send Check Does

The pre-send check hook performs three critical validation steps on every generated response:

1. Analyzes draft output – Scans text for overly long sentences, dense jargon, or irrelevant tangents that might overwhelm users with ADHD.
2. Applies ADHD-friendly style rules – Automatically breaks complex ideas into bullet points, inserts short summaries, or restructures paragraphs for better cognitive load management.
3. Blocks or rewrites – Aborts the send operation if content exceeds configurable thresholds (such as character limits or distraction keyword density) and requests a revised, more focused reply from the LLM.

Architecture and Key Files

The pre-send check functionality is distributed across several specific files in the repository:

• skills/i-have-adhd/SKILL.md – The skill manifest that registers the pre_send_check hook with the Cursor runtime.
• skills/i-have-adhd/agents/openai.yaml – Contains the prompt template and hook wiring for OpenAI backend implementations.
• skills/i-have-adhd/agents/gemini.toml – Houses the equivalent prompt template for the Gemini LLM backend.
• hooks/hooks.json – Stores runtime configuration including toggle switches, length limits, and keyword filters.
• scripts/run_evals.py – Command-line utility that executes the skill and automatically triggers the pre-send check during evaluation.

When a response is generated, the Cursor runtime invokes the pre_send_check hook defined in the skill manifest, which reads its rules from hooks/hooks.json and applies the logic specified in the respective agent configuration file.

Configuring the Pre-Send Check

You can customize the validation behavior by editing the JSON configuration at hooks/hooks.json. The following settings control how strictly the hook filters content:

• enabled – Boolean flag that turns the pre-send check on or off (default: true).
• max_length – Maximum character count allowed for a single response (default: approximately 500 characters).
• bullet_threshold – Minimum number of distinct points the hook identifies before forcing a bullet-list conversion.
• distraction_keywords – Array of trigger words (such as "actually", "maybe", or "basically") that, when detected, prompt an automatic rewrite.

Changes to this file take effect immediately on the next invocation without requiring a restart.

Practical Implementation Examples

Running the Skill from the Command Line

To execute the skill manually and observe the pre-send check in action, run the evaluation script from the repository root:

python -m scripts.run_evals --skill i-have-adhd

This command loads the skill, generates a response through the configured LLM, and automatically applies the pre_send_check validation before displaying any output to the terminal.

Integrating with the Cursor Client

When using the skill within a Cursor environment, the pre-send check runs transparently during the ask method call:

from cursor import CursorClient

client = CursorClient()
response = client.ask(
    "Explain the concept of neurodiversity in three sentences.",
    skill="i-have-adhd"
)

print(response)  # Output has already passed the pre-send check filters

Modifying Hook Configuration

To customize validation parameters, edit hooks/hooks.json with your preferred constraints:

{
  "pre_send_check": {
    "enabled": true,
    "max_length": 400,
    "bullet_threshold": 2,
    "distraction_keywords": ["actually", "maybe", "basically"]
  }
}

Disabling the Check Temporarily

For queries requiring detailed technical responses, you can bypass the hook by passing an options flag:

response = client.ask(
    "Give me a long technical description of quantum computing.",
    skill="i-have-adhd",
    options={"pre_send_check": false}
)

Summary

• The i-have-adhd skill provides a pre_send_check hook that validates AI responses before they reach the user interface.
• Configuration resides in hooks/hooks.json, supporting custom length limits, keyword filters, and enable/disable toggles.
• Implementation logic is split between agents/openai.yaml and agents/gemini.toml depending on your LLM backend.
• You can invoke the check automatically via scripts/run_evals.py or programmatically through the CursorClient interface.

Frequently Asked Questions

What triggers the pre-send check to rewrite a response?

The hook triggers a rewrite when content exceeds the max_length threshold, contains words listed in distraction_keywords, or fails the structural complexity analysis defined in the agent prompt templates.

Can I disable the pre-send check for specific queries?

Yes. Pass options={"pre_send_check": false} in the client.ask() method call, or set "enabled": false in hooks/hooks.json to disable it globally.

Where is the pre-send check logic defined?

The primary logic resides in skills/i-have-adhd/SKILL.md where the hook is registered, with backend-specific implementations in skills/i-have-adhd/agents/openai.yaml and skills/i-have-adhd/agents/gemini.toml.

How do I adjust the length limits for ADHD-friendly outputs?

Edit the max_length value in hooks/hooks.json to set your preferred character limit. The default configuration limits responses to approximately 500 characters, but you can reduce this to 300 or 400 for more concise interactions.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}