# I Have ADHD: the Next Action Should Not Be Hidden in Paragraph Seven

Date: 2026-10-15

Both responses below answer the same question about a failing authentication test. Only one can be acted on without reading it twice.

**Before** *(illustrative—not from a real project)*

> Great question! JWT failures have several causes: signing algorithm, secret key, expiry settings, or middleware placement. One approach: add logging around the verify call. Another: inspect the payload with jwt.io. Worth checking whether the test environment loads the right variables—secrets often aren't exported in CI. By the way, your expiry settings may also differ between environments. Hope that helps!

**After** *(illustrative — `auth.service.ts` is a fictional filename)*

> Edit `auth.service.ts` line 47: `jwt.verify` is missing the `algorithms` option and accepts any algorithm by default in this version.
>
> 1. Open `auth.service.ts`
> 2. Change `jwt.verify(token, secret)` to `jwt.verify(token, secret, { algorithms: ['HS256'] })`
> 3. Run `npm test -- auth.test.ts`
>
> Next: paste the first failing line if any test still fails.

The first response offers preamble, three alternative paths, a tangent, and a closer. The second leads with the cause, sequences three steps, and ends with one action. The diagnosis is present in both. The first adds reading time, not information.

<!--more-->

I Have ADHD is a skill file (a prompt-based instruction installed into Claude Code, Codex, or other coding agents) that applies ten documented output rules to every response in a session until the user turns it off.

## The rules

The `SKILL.md` names the cognitive logic: "Knowing the answer is not doing the answer. The friction between 'got it' and 'done it' is where work dies." The rules address that friction directly.

**Rule 1 — Lead with the next action.** The first line is something the reader can do: a command, a path, a snippet. Context follows, if needed.

**Rule 2 — Number multi-step tasks.** Each step is one bounded action. The fewest steps that work beats the most complete path abandoned.

**Rule 3 — End with one concrete next step.** One thing doable in under two minutes. "Next: run `npm test` and paste the first failing line" rather than "Hope that helps!"

**Rule 4 — Suppress tangents.** Finish the first issue, then offer the second as a separate question. "By the way" becomes a follow-up offer, not a detour.

**Rule 5 — Restate state every turn.** The reader cannot hold "step 3 of 5" between messages. The skill restates it: *"Step 3 of 5 done: schema updated. Next: run the backfill script."*

**Rule 9 — Cap lists at five items.** This shapes what is displayed, not what is retained. Relevant items are kept and shown when asked or when they become the next items to address. Analysis and search results are not capped—only presentation.

Four more rules: specific time estimates in concrete units (Rule 6); completed work in concrete terms, not buried in a recap (Rule 7); errors as cause and fix, no "uh oh" (Rule 8); no preamble, no recap, no closing pleasantries (Rule 10).

The repository credits *The Adult ADHD Tool Kit* by Ramsay and Rostain, adapted for LLM output rather than personal organization.

## Positioning

The repository's own tagline is "No ADHD diagnosis needed." Output that leads with the action and suppresses tangents can be useful to any reader working through a multi-step session under time pressure—developers troubleshooting, executives reading briefings structured as walls of options.

Individual preferences for information density differ. The skill is a particular format, not a universal improvement.

## When the rules yield

The `SKILL.md` documents explicit overrides. A destructive action—force push, schema migration, dropping a table—requires confirmation, regardless of brevity. Three consecutive failed turns call for naming the wrong assumption, not iterating further. Real ambiguity gets one clarifying question rather than a guess.

Concision that removes a necessary risk warning is not output discipline—it is a gap in the answer.

## The anomaly in the argument

Every other skill in this series addresses an engineering-specific problem. This one is different. Action-first output, numbered steps, suppressed tangents, one concrete next step—these describe how most people prefer to receive instructions, not only developers.

The skill still requires an active agent session. But a broadly useful format emerged from a developer-focused ecosystem—because that community treats clarity and actionability as engineering properties worth encoding.

## The next problem

Readable output matters. For security findings, readability without evidence is a liability. The next post examines a skill designed to verify findings independently before they reach a report.

---

*Source: github.com/ayghri/i-have-adhd — SKILL.md and README, retrieved 8 October 2026.*

#AgentSkills #DeveloperTools #SoftwareEngineering