How to Suppress Tangents While Maintaining Context in i-have-adhd

The i-have-adhd repository implements tangent suppression through Rule 4 in its SKILL.md file, which instructs the LLM to finish the primary response before offering secondary issues as separate follow-up questions.

The ayghri/i-have-adhd project provides a structured skill for AI agents designed to streamline communication patterns associated with ADHD. By enforcing explicit rules for response generation, this open-source tool ensures focused, context-aware interactions without permanently losing track of secondary concerns. Understanding how to suppress tangents while maintaining context in i-have-adhd requires examining the rule-based architecture defined in the skill's metadata and agent configuration files.

The Rule-Based Architecture of i-have-adhd

The skill operates on a set of explicit behavioral rules that shape every response generation cycle. These rules are not enforced through traditional code logic but through prompt engineering embedded in the skill's definition files.

Rule 4: The Core Tangent Suppression Mechanism

The primary technique for suppressing tangents is defined in Rule 4 within skills/i-have-adhd/SKILL.md at lines 66-73:

"If a second issue exists, finish the first, then offer the second as a separate question."

According to the source code in SKILL.md §4, a tangent is avoided by completing the current answer first, then optionally presenting the next issue as a separate prompt. This rule transforms how the LLM structures multi-part responses, preventing the interleaving of unrelated tasks.

Technical Implementation Details

The tangent suppression behavior relies on a declarative configuration approach rather than procedural code. When the skill is invoked via /i-have-adhd or $i-have-adhd, the response generator consults these rules before emitting output.

Skill Metadata Configuration

The skills/i-have-adhd/SKILL.md file serves as the central authority for behavior definition. The hosting plugin (whether Claude, Codex, or similar) reads this file when loading the skill and injects the rule text directly into the system prompt fed to the LLM. This ensures the model enforces the "finish first, offer second" pattern without requiring runtime detection algorithms.

Agent Definition and Runtime Behavior

The skills/i-have-adhd/agents/openai.yaml file declares the skill's display name and default prompt configuration. This agent definition tells the host platform to apply the i-have-adhd rules whenever the skill is active. During runtime, when the LLM detects a secondary issue while processing a request, it defers that portion until the primary answer is completed, then offers it as a new question using the format: "Separately: [issue description]. Want me to handle that next?"

Practical Usage Examples

Implementing the tangent suppression technique requires activating the skill through your agent's plugin system.

Installing and Invoking the Skill

# Example: Declaring the skill in a Claude plugin manifest

plugins:
  - name: i-have-adhd
    path: .claude-plugin
    invoke: "/i-have-adhd"

# Using the skill with Claude Code (CLI)

claude plugin marketplace add ayghri/i-have-adhd
claude plugin install i-have-adhd@i-have-adhd

# Then type:

/i-have-adhd

Typical LLM Output Respecting Rule 4

When the skill is active, responses follow a strict separation pattern:

Answer

1. Run npm install jsonwebtoken@latest.
2. Edit src/auth.ts at line 42.
3. Run npm test -- auth.spec.ts.

Separately: There is also a stale dependency in package.json. Want me to update that next?

In this example, the secondary issue (the stale dependency) is not interleaved with the primary fix. Instead, it appears after the primary steps conclude, exactly as Rule 4 dictates in the source configuration.

Key Files Supporting Tangent Suppression

Several files work together to implement this behavior:

• skills/i-have-adhd/SKILL.md – Contains the complete rule set, including Rule 4 (Suppress tangents) at lines 66-73
• skills/i-have-adhd/agents/openai.yaml – Agent definition enabling the skill for OpenAI-based hosts
• README.md – High-level description of the skill's purpose and rule philosophy
• plugin.json – Root metadata used by the plugin system to load and initialize the skill
• INSTALL.md – Platform-specific installation instructions for Claude, Codex, and other agents

Summary

• Rule 4 in SKILL.md (lines 66-73) provides the explicit instruction to finish primary responses before addressing secondary issues.
• The skill uses declarative prompt engineering rather than code-based detection to suppress tangents.
• Secondary issues are deferred and offered as separate questions using the "Separately:" format.
• The openai.yaml agent definition activates these rules for compatible LLM hosts.
• Invocation occurs through /i-have-adhd or $i-have-adhd commands depending on the platform.

Frequently Asked Questions

How does i-have-adhd detect tangents programmatically?

The system does not use programmatic detection. Instead, the LLM receives the rule text from SKILL.md as part of its system prompt, allowing it to recognize potential tangents during generation and restructure the output accordingly. This approach relies on the model's ability to follow instructions rather than traditional code logic.

Can I customize Rule 4 to allow certain types of tangents?

Yes, since the rules are defined in the markdown file skills/i-have-adhd/SKILL.md, you can modify the text at lines 66-73 to adjust the tangent suppression behavior. However, changing this file affects all responses generated under this skill, so modifications should maintain the structural integrity expected by the agent definition.

What platforms support the i-have-adhd skill?

According to the repository structure, the skill supports Claude and OpenAI-based agents through the configuration files in skills/i-have-adhd/agents/. The INSTALL.md file provides specific instructions for various hosting platforms, indicating broad compatibility with modern LLM agent frameworks.

How does the skill maintain context when suppressing tangents?

The skill preserves context by explicitly acknowledging secondary issues rather than ignoring them. By offering the deferred topic as a separate question (e.g., "Want me to handle that next?"), the LLM maintains awareness of the pending task while keeping the immediate response focused. This technique satisfies the "suppress tangents" requirement without losing track of important secondary concerns.

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