How the i-have-adhd Plugin Estimates Time for ADHD-Friendly Responses

The plugin does not calculate time itself—it injects a behavioral rule into the system prompt that instructs the language model to "ballpark in concrete units."

The i-have-adhd plugin helps developers with ADHD by restructuring AI responses for clarity and actionability. One of its most practical features is the automatic inclusion of concrete time estimates (e.g., "About 15 minutes" or "roughly an hour"). This article explains how the plugin enables time estimation behavior based on the rules defined in its source code.

How Time Estimation Actually Works

The plugin does not perform any deterministic calculation of task duration. Instead, it relies on the underlying language model's knowledge and reasoning capabilities, guided by an explicit rule in the skill definition.

The Rule That Drives Time Estimates

In skills/i-have-adhd/SKILL.md, Rule 6 — "Give specific time estimates" provides the instruction:

- **Give specific time estimates**  
  Ballpark in concrete units (e.g., "About 15 minutes...", "roughly an hour...") instead of "a while".

This rule appears at lines 82–88 of the SKILL.md file. The directive is unambiguous: replace vague temporal language with quantified, concrete estimates.

Plugin Architecture: Prompt Injection, Not Computation

The plugin's role is strictly orchestration. When activated, the code in .opencode/plugins/i-have-adhd.mjs performs these steps:

1. Reads the SKILL.md file
2. Strips YAML front-matter from the markdown
3. Injects the raw ruleset into the system prompt

The relevant implementation spans lines 35–42 and 55–70:

// From .opencode/plugins/i-have-adhd.mjs
// The plugin loads SKILL.md and processes it for injection
async function loadSkill() {
  const skillPath = path.join(__dirname, '../../skills/i-have-adhd/SKILL.md');
  const content = await fs.readFile(skillPath, 'utf8');
  // Strip frontmatter and return body
  return content.replace(/^---[\s\S]*?---/, '').trim();
}

// Later, the ruleset is prepended to every system prompt
function injectSkill(systemPrompt) {
  return `${skillRules}\n\n${systemPrompt}`;
}

Once injected, the model receives the full rule set as context and generates time estimates autonomously based on:

• The complexity of the described task
• Its training knowledge about typical development workflows
• The explicit stylistic constraint to use concrete units

Activating Time Estimates in Your Session

The ADHD-friendly time estimates appear automatically once the skill is loaded. You have two activation modes.

On-Demand Activation

Trigger the skill for a single session by sending the command:

await chat.sendMessage('/i-have-adhd');

Always-On Activation

For persistent behavior across turns, create the flag file:

touch ~/.config/opencode/.i-have-adhd-always

The hooks/always-on.mjs runtime component monitors this flag and ensures the skill ruleset—including Rule 6—remains active for every interaction.

What the Output Looks Like

With the skill active, model responses transform from vague to actionable. A typical reply includes time estimates inline:

Run `npm install jsonwebtoken`, then edit `src/auth.ts:42`.

About **15 minutes** if tests already cover the change; an **hour** if they don't.

Notice the ** dual structure**: primary estimate with a conditional branch. This mirrors the rule's intent to provide concrete, scenario-aware time framing.

Why This Architecture Matters

The plugin's design reflects a key insight: time estimation for creative/development work is inherently uncertain. Rather than bake in brittle heuristics, the plugin:

• Leverages model reasoning — the LLM assesses task complexity holistically
• Remains maintainable — rule changes require only editing SKILL.md
• Preserves flexibility — estimates adapt to context rather than following rigid formulas

According to the ayghri/i-have-adhd source code, no timing logic exists in the plugin's JavaScript files. The entire behavior emerges from prompt engineering and model compliance with the declared ruleset.

Key Files for Time Estimation Behavior

File	Function
skills/i-have-adhd/SKILL.md	Contains Rule 6 that mandates concrete time units; authoritative source of the behavior
.opencode/plugins/i-have-adhd.mjs	Loads and injects SKILL.md content into system prompts (lines 35–42, 55–70)
hooks/always-on.mjs	Runtime persistence mechanism for the always-on flag

Summary

• Time estimation is model-driven, not computed by the plugin
• Rule 6 in SKILL.md explicitly instructs the model to use concrete time units
• The plugin only handles prompt injection via .opencode/plugins/i-have-adhd.mjs
• Activation modes: /i-have-adhd command or ~/.config/opencode/.i-have-adhd-always flag
• No fixed algorithm — estimates adapt to task context through LLM reasoning

Frequently Asked Questions

Does the plugin calculate how long tasks take?

No. The plugin contains no timing logic, heuristics, or lookup tables. It solely injects Rule 6 into the system prompt, which instructs the language model to provide its own time estimates based on context and training knowledge.

Can I customize the time estimation behavior?

Yes—by editing skills/i-have-adhd/SKILL.md. Modify Rule 6's wording to change how estimates are phrased, or adjust other rules to complement the timing guidance. Changes take effect immediately on plugin reload; no code changes required.

Why use ballpark estimates instead of precise calculations?

Precise software development time estimates are notoriously unreliable. The "ballpark in concrete units" approach from Rule 6 acknowledges uncertainty while still providing actionable framing—critical for ADHD users who benefit from temporal boundaries without false precision.

What happens if the model ignores the time estimate rule?

The ruleset injection in .opencode/plugins/i-have-adhd.mjs places the instructions at the start of the system prompt, maximizing compliance likelihood. If estimates still fail to appear, verify the plugin activated correctly: check for the /i-have-adhd command response or confirm the always-on flag file exists.

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