Your coding agent's new anti-waffle filter

What it does

This is a Claude Code and Codex skill that restructures how the agent replies. Instead of preamble-heavy, meandering responses, it forces action-first output: numbered steps, concrete next actions, and no ceremonial sign-offs. The README shows a stark before/after where a rambling auth-debugging monologue collapses into three lines and a prompt to paste the first failing test line.

The interesting bit

The ten rules are drawn from actual ADHD clinical literature—specifically The Adult ADHD Tool Kit—but repurposed as system prompts for an LLM. That makes it a rare case where productivity psychology is translated directly into prompt engineering, not just time-management advice for humans.

Key highlights

• Ten explicit output rules, including “Lead with the next action,” “Cap lists at 5 items,” and “Specific time estimates (minutes, not ‘a bit’).”
• Can be invoked explicitly or, in Codex, triggered implicitly when the task suits action-first formatting.
• Rules are stored in an editable SKILL.md file; tweak and re-invoke without reinstalling.
• Before/after comparison in the README demonstrates a dramatic reduction in verbal overhead for a typical debugging workflow.

Caveats

• Only works within Claude Code and Codex; not a general-purpose prompt wrapper for other agents or APIs.
• Implicit triggering is only documented for Codex; Claude Code users must invoke the skill explicitly.

Verdict

Grab it if you spend more time scrolling past LLM throat-clearing than reading answers. Skip it if you already have your own system-prompt scaffolding or prefer conversational, exploratory agent replies.