Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 

README.md

adhd

A Claude Code plugin that shapes and restructures the assistant's output for a reader with ADHD, and anyone who wants action-first, low-friction, digestible responses. Two skills, one concern: arrange output so an ADHD brain can act on it. shape sets a standing house style; clarify rescues one specific artifact.

Skill What it does
/adhd:shape Shape responses to lead with the next action, number multi-step work, restate state, cap and rank lists, estimate time concretely, make wins visible, and cut preamble, recap, and closers
/adhd:clarify Faithfully restructure a dense, decision-heavy message already on screen. Chunk it one-decision-at-a-time, define the session's jargon, and surface what you must decide; renders an HTML decision table for big content

What it does

/adhd:shape, set the house style

The skill re-anchors ten output rules grounded in five facts about how an ADHD brain reads: working memory is small, knowing is not doing, starting is the hardest step, time reads as uniform, and dopamine is scarce. It shapes output to lead with a concrete next action, number multi-step work, restate state across turns, cap and rank lists at five, give time estimates in real units, make finished work visible, keep an error tone flat, and drop preamble, recap, and closing pleasantries.

It knows when not to compress: an "explain / walk me through" request runs full-length, a destructive action gets a confirmation first (safety over brevity), a debug spiral pauses to name the wrong assumption, and a genuinely ambiguous request earns one clarifying question.

/adhd:clarify, rescue one dense artifact

Where shape governs how the assistant writes going forward, clarify acts once on something already on screen: a wall-of-text interview round, a jargon-thick design memo, a recommendation you have to re-read three times. It restructures that exact artifact. One decision per chunk, a glossary of the session's own shorthand, and the actual choices pulled to the surface, faithfully. The move is restructure, never simplify: precision and reading level stay fixed; only the arrangement changes (lowering the altitude is education:explain's job and turning it into a visual explainer is education:eli5's, both deliberately disjoint concerns). Four hard fidelity rules keep a clarification of a decision document from corrupting the decisions: operative terms quoted verbatim, original item numbers kept as back-links, omissions named explicitly, and a closing line that the clarification is a lens. Final answers are validated against the original text. For big or decision-dense content it renders an HTML decision table (item, recommendation, alternative, and what you're deciding, with rows numbered so a terminal answer maps back), honoring the Artifact tool contract and degrading to a local HTML file, then structured terminal markdown, where the Artifact surface is unavailable.

Triggering: on-demand, session-standing once invoked

The skill is on-demand by design. It does not auto-fire on every message. It surfaces when you ask for it in plain language ("ADHD-friendly", "action-first", "give me the structured version", "cut the preamble") or when you invoke /adhd:shape directly.

Once invoked, shape's rules persist as a standing instruction for the rest of the session. Invoke it once at the start of a session and every following response is shaped, no need to repeat it. Invoke it whenever you want that output shape; skip it when you don't. Turn it off mid-session with "stop shaping" or "normal output".

One durability caveat: the standing posture is content-based persistence, so context compaction/summarization in a long session can erode it. If responses stop being shaped after a compaction, re-invoke /adhd:shape.

clarify is on-demand too, but one-shot, not session-standing: it surfaces on a plain-language cue ("make this clear", "clarify this", "help me digest this", "break this down", "I can't parse this", "what am I actually deciding") or a direct /adhd:clarify, acts on one artifact, and changes nothing about how later responses are written. Its triggers are kept disjoint from education:explain's comprehension cues ("I don't get it", "explain simply") and from education:eli5's picture cues ("ELI5", "draw me this") so the auto-firing skills route on intent, restructure faithfully vs drop the altitude vs change the medium, rather than colliding on their shared "previous response" default target.

Deferred: deterministic zero-invocation always-on

Arming the shaping at every session start with no invocation at all is a deliberate non-goal for this version, recorded here so it is not re-litigated:

  • Why it is not a userConfig switch. A userConfig boolean substitutes only into an already-invoked skill's body; it cannot flip frontmatter or cause a skill to auto-invoke. So no config toggle can, by itself, turn on always-on. A switch that cannot act would be a dead knob.
  • The only mechanism that delivers it is a hook. Deterministic every-session arming needs a SessionStart (or UserPromptSubmit) hook that injects the rules as additional context. That is a code-execution trust surface, adopted only on demonstrated need, and the session-standing behavior above already covers the common case (invoke once, shaped for the rest of the session).
  • Trigger to build it. Demonstrated demand for zero-invocation auto-arm. When that lands, the hook ships off by default (opt-in), so enabling the plugin never silently rewrites every response.

Mutual exclusion with terse-for-tokens output shapers

Do not run this alongside a token-minimizing output shaper such as caveman at the same time. They pull in opposite directions on the same axis, the shape of the assistant's output:

adhd:shape caveman
Objective Add structure and cues for reader accessibility Strip words to save tokens
For The human reading the output The token budget

Two output-shape disciplines active at once produce a contradictory, unpredictable mix. Pick one for a given session. (There is no runtime coupling between the plugins to enforce this. It is a usage guideline.)

Install

/plugin marketplace add melodic-software/claude-code-plugins
/plugin install adhd@melodic-software

Configuration

None. The plugin is zero-config and zero-prerequisite. No userConfig, no setup skill, no external tools. Enable it and invoke /adhd:shape or /adhd:clarify.

Attribution

/adhd:shape is reauthored, not forked, from ayghri/i-have-adhd (MIT): the ten rules' substance is preserved, the wrapper is adapted to this marketplace's discovery discipline (no auto-fire-on-any-message), and the prose is rewritten. The underlying communication strategies adapt The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain from personal organization to how an assistant shapes its output. /adhd:clarify is original to this plugin.

License

MIT. Because this is a derivative reauthor, the plugin ships its own LICENSE retaining the upstream copyright notice (Ayoub Ghriss) alongside Melodic Software's, per the MIT requirement that the original copyright and permission notice travel with substantial portions of the work.