How the i-have-adhd Skill Ensures Every Response Ends with a Concrete Short Next Action

The i-have-adhd skill enforces a strict "end-with-one-concrete-next-action" rule by selecting a single bounded task, placing it as the final line prefixed with "Next:", and running a pre-send check that strips trailing filler sentences.

This behavioral design pattern, implemented in the ayghri/i-have-adhd repository, addresses core challenges faced by users with ADHD: decision paralysis, task ambiguity, and cognitive overload. The skill transforms every AI response into a clear execution sequence with an unmistakable final step.

The Rule: One Concrete Action Under Two Minutes

The canonical skill definition in skills/i-have-adhd/SKILL.md codifies this behavior as Rule 3 (lines 57‑63). The guideline reads:

"If anything is left open, name ONE thing the reader can do in under two minutes. Even 'open the file' counts."

Here is the relevant excerpt from the skill definition:

### Rule 3: End with one concrete short next action

After answering, if anything is left open, name ONE thing the reader can do in under two minutes. Even "open the file" counts.

This goes at the very end. No "anything else?" after it.

This rule hard-codes three constraints directly into the response generation logic:

• Singularity: Exactly one action, never multiple options
• Boundedness: Completion time ≤ 2 minutes
• Terminal position: Final line with no subsequent commentary

The Four-Step Enforcement Pipeline

The skill implements Rule 3 through a deterministic pipeline in the response generator:

Step 1: Identify Unfinished Work

After generating all required actions for the current turn, the skill performs an inventory of remaining open tasks. This check runs against the current conversation context and any deferred work from previous exchanges.

Step 2: Select a Single Bounded Action

The generator filters candidate actions through the two-minute threshold. Valid selections include atomic operations like:

• "run the test"
• "open the file"
• "paste the error"
• "check line 47"

Multi-step or ambiguous actions are rejected or broken down further.

Step 3: Render as Terminal Line with "Next:" Prefix

The selected action is formatted as the absolute final line, prefixed with "Next:" (or equivalent cue). No additionaltext follows. This creates a visual anchor that captures remaining attention and eliminates scroll ambiguity.

Example output structure:

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

1. Open `src/auth.ts`
2. Replace `verifyToken` (lines 42‑58) with the snippet below
3. Run `npm test -- auth.spec.ts`

Next: run `npm test` and paste the first failing line.

In this example:

• First line provides immediate context
• Numbered list breaks multi-step work into bounded chunks
• Terminal "Next:" line satisfies Rule 3 with a single executable instruction

Step 4: Pre-Send Sanity Check

Before transmission, the skill executes a validation routine defined in SKILL.md lines 28‑38. This pre-send check performs automatic cleanup:

• Removes trailing sentences asking "anything else?"
• Strips open-ended offers like "let me know if you need more help"
• Verifies the "Next:" line exists and is terminal

The check guarantees no dilution of the final action's impact.

Short Response Variant

For simpler queries, the structure compresses but preserves the rule:

Run `git pull` to fetch the latest changes.

Next: verify the merge with `git status`.

Even with minimal content, the concrete short next action appears as the terminal element.

Implementation Files

File	Purpose	Key Lines
skills/i-have-adhd/SKILL.md	Canonical rule definition	Lines 57‑63 (Rule 3), lines 28‑38 (pre-send check)
.cursor/skills/i-have-adhd/SKILL.md	Cursor IDE compatibility mirror	Synced with canonical version
hooks/always-on.mjs	Runtime enforcement hook	Applies rules to all generated responses

The always-on.mjs hook ensures these constraints apply universally across supported runtimes, making the concrete short next action guarantee non-optional.

Design Rationale

The two-minute threshold specifically targets activation energy reduction. For users with ADHD, identifying "what do I do right now" often presents greater friction than executing the task itself. By pre-selecting and isolating a single micro-action, the skill eliminates:

• Decision fatigue from multiple valid options
• Overwhelm from underspecified or open-ended instructions
• Delay from uncertain starting points

Summary

• Rule 3 in skills/i-have-adhd/SKILL.md mandates exactly one actionable final step
• Actions must complete in ≤ 2 minutes and appear as the terminal line
• The "Next:" prefix creates visual salience
• Pre-send checks (lines 28‑38) strip diluting commentary
• The always-on.mjs hook enforces these constraints across all responses

Frequently Asked Questions

What happens if there's no unfinished work?

The skill still generates a concrete short next action focused on verification or progression, such as "Next: confirm the output matches your expected result." The Rule 3 requirement applies universally, not conditionally on open tasks.

Why two minutes specifically?

The two-minute threshold balances atomic executability with meaningful progress. It exceeds typical "next-action" recommendations (60–90 seconds) to accommodate brief file inspections or single command executions, while remaining below the threshold where task initiation anxiety typically rises.

Can the "Next:" prefix be customized?

No. The prefix is hard-coded in the skill definition to maintain consistency across all interfaces. Standardization ensures users develop predictable scanning patterns—their eyes automatically seek the terminal "Next:" line regardless of response complexity.

Does this rule ever conflict with other skill behaviors?

The pre-send check (lines 28‑38) resolves conflicts by prioritizing Rule 3. If another guideline suggests trailing pleasantries or follow-up prompts, the sanity check removes them before transmission. The concrete short next action always wins.

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