How the i-have-adhd Suppress Tangents Feature Works: Cursor Skill Rule 4

The i-have-adhd suppress tangents feature forces LLMs to complete one issue before addressing another, converting side remarks into separate follow-up questions via a rule-driven post-processing step defined in SKILL.md.

The i-have-adhd repository provides a Cursor skill that shapes AI responses for readers with ADHD. Its suppress tangents functionality ensures that every output remains laser-focused on the primary query by stripping unrelated commentary and enforcing a strict "one issue at a time" structure.

What the Suppress Tangents Rule Does

Rule 4: Suppress Tangents operates as a behavioral constraint on the LLM's output generation. According to the skill definition in skills/i-have-adhd/SKILL.md (lines 66-73), the rule mandates three specific behaviors:

• Finish the first issue completely before mentioning any secondary topics.
• Present secondary issues as separate questions only after the primary answer concludes.
• Eliminate "by-the-way" remarks such as phrases starting with "Also, ..." or "By the way, ..." during the pre-send validation phase.

This ensures that readers receive a single, actionable response without cognitive load from tangential information.

How the Feature is Enforced

The suppress tangents behavior is not implemented through standalone code but through rule-driven post-processing embedded in the skill specification. When the skill is active, the Cursor skill engine executes a three-step enforcement pipeline:

1. Parse the skill definition from .cursor/skills/i-have-adhd/SKILL.md, loading the "Suppress tangents" constraints into the response generator.
2. Run the pre-send filter, which deletes any text matching tangent patterns—specifically secondary issues appended after the main answer or conversational asides.
3. Validate the final output against structural requirements: the first line must contain the actionable step, and the last line must state the next concrete action. If tangential content remains, the engine discards the response and regenerates it.

This pipeline guarantees that every response adheres to the ADHD-friendly format while the skill is active.

Implementation Files and Configuration

The feature is defined across multiple files in the repository:

• skills/i-have-adhd/SKILL.md and .cursor/skills/i-have-adhd/SKILL.md — Both contain the full skill definition, including the "Suppress tangents" rule at lines 66-73.
• skills/i-have-adhd/agents/openai.yaml — Example agent configuration that loads the skill using the {{< skill i-have-adhd >}} directive.
• skills/i-have-adhd/agents/gemini.toml — Alternative configuration for Gemini models that applies the same constraint set.

These markdown and YAML files collectively define the behavior without requiring compiled code or API modifications.

Practical Examples

The following examples demonstrate how the rule transforms LLM output during the pre-send check.

Before applying Rule 4 (tangential output):

Here's the fix for the login bug.
By the way, your README is missing a section on environment variables.
Do you want me to add that now?

After the suppress tangents rule is applied:

Here's the fix for the login bug.
Separately: your README is missing a section on environment variables. Want me to handle that next?

Loading the skill in an agent configuration:

# agents/openai.yaml

system: |
  {{< skill i-have-adhd >}}
user: |
  How do I refactor this function?

When the model generates a response, the engine automatically enforces Rule 4, stripping extra commentary and isolating secondary issues as distinct questions.

Summary

• The i-have-adhd suppress tangents feature is defined as Rule 4 in the skill's markdown specification, not through standalone code.
• Located in skills/i-have-adhd/SKILL.md (lines 66-73), the rule forces the LLM to complete one issue before introducing another.
• Secondary topics must appear only as separate follow-up questions, and "by-the-way" phrases are removed during the pre-send validation step.
• The Cursor skill engine implements this through post-processing filters that validate output structure and regenerate responses containing tangential content.

Frequently Asked Questions

Where is the suppress tangents rule defined in the repository?

The rule is explicitly defined in the "4. Suppress tangents" section of skills/i-have-adhd/SKILL.md at lines 66-73. An identical copy exists at .cursor/skills/i-have-adhd/SKILL.md to ensure compatibility with the Cursor skill engine's file resolution.

Does this feature require changes to the underlying LLM?

No. The suppress tangents feature operates entirely through post-processing and prompt engineering within the Cursor skill framework. The skill engine intercepts the generated response, applies the filter rules, and regenerates content if necessary—no fine-tuning or model weights are modified.

How does the pre-send check determine if content is tangential?

The pre-send validator scans for structural and lexical patterns indicating deviation from the primary topic. It specifically targets phrases like "Also," or "By the way," and validates that the first line contains the actionable answer while the last line contains the next concrete step. Content violating this linear structure is flagged as tangential and removed.

Can I use this skill with AI agents outside of Cursor?

The skill is designed for the Cursor skill engine and uses Cursor-specific syntax like {{< skill i-have-adhd >}}. While the markdown rules in SKILL.md could theoretically be adapted for other frameworks, the automated pre-send filtering and validation require Cursor's native skill engine to enforce the suppress tangents behavior automatically.

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