# akhouri-adhd-ideation-divergente-parallele-2026-07-20

## Veille

Udit Akhouri releases **ADHD**, an open source (MIT) skill for "parallel divergent ideation" for coding agents: N **isolated** agent calls under deliberately distorted cognitive frames, then a separate critic that scores, clusters, **flags traps**, and deepens the survivors — an **architectural** fix (not a prompt) to LLM premature convergence.

## Titre Article

ADHD — a skill for agents (Parallel Divergent Ideation for Coding Agents)

## Date

2026-07-20

## URL

https://github.com/UditAkhourii/adhd

## Keywords

ADHD, Udit Akhouri, parallel divergent ideation, premature convergence, cognitive anchoring, Tree-of-Thought, Chain-of-Thought, cognitive frames, generator vs critic, trap detection, skills, parallel agents, npx skills add, rage-quit button, Claude Agent SDK, LLM judge eval, multi-agent brainstorming, repowire

## Authors

Udit Akhouri (@akhouriudit)

## Ton

**Profile:** Open source project README | Author-promoter | Demonstrative and instrumented (before/after benchmark, committed transcripts) | ⚠️ self-reported evaluations: problems chosen by the author, scoring by LLM judge — to be treated as claims, not independent measurements.

## Pense-betes

- **Architectural thesis**: anchoring is not a prompting problem — CoT anchors on its first output, and even Tree-of-Thought "widens the search but traverses a single shared context, so the anchoring persists across branches."
- **2-phase watertight mechanics**: *Diverge* = N isolated agent calls (zero shared context), each under an imposed cognitive frame + a system prompt that **forbids evaluating**; *Focus* = a separate critic (opposite system prompt) scores originality/viability/fit, clusters, **flags traps with reasons**, deepens the top-K.
- **15 cognitive frames** shipped + selection logic + custom frames; triggers on "design decisions, fuzzy debugging, naming, API surface design, strategy" and any prompt shaped like "give me a few ways to…".
- **Signature demo (CLI that freezes for 90s)**: baseline = 4 textbook patterns (progressive timeouts, backoff, hedging, streaming); ADHD = 30+ ideas across 6 clusters, 20 named traps, and the non-obvious pick **"rage-quit"** — a button that heats up with the wait and reroutes the request to a faster/cheaper model ("the slow model might just be the wrong model").
- **Author evaluations (6 open problems, LLM judge)**: breadth 9.00 vs 4.83; novelty 7.83 vs 2.67; **trap detection 9.50 vs 1.83**; actionability 9.50 vs 6.50 — ADHD wins 5/6.
- **Skills ecosystem distribution**: `npx skills add UditAkhourii/adhd` (~50 auto-detected agents: Claude Code, Cursor, Codex, Cline, Gemini CLI, Windsurf…), `/adhd "…"`, CLI/npm lib `adhd-agent`; built on the Claude & Codex Agent SDKs.
- **Real traction**: ~1k stars, a feature on The New Stack, a preprint; adopted by repowire (PR #313, frames → "frame-shifted temp peers" of the mesh-orchestrator), mstack (`think` plugin), zk-flow-oss, and an **independent research review** (testdouble/han, 11 sources, 8 rounds) whose findings land in public issues #16-18.
- **Telling engineering detail**: the SKILL.md keeps its description on a single line ≤600 characters because some Codex builds truncate or reject multiline YAML.

## RésuméDe400mots

Udit Akhouri releases **ADHD** ("a skill for agents"), an open source project (MIT, v0.1.4, ~1,000 stars) that tackles the **premature convergence** of autoregressive reasoning: an LLM anchors on its first idea, and tree-based methods don't really escape it — "Tree-of-Thought widens the search but traverses a single shared context, so the anchoring persists across branches." The project's stance: this is an **architecture problem, not a prompting problem**.

The mechanics consist of two **watertight** phases. *Diverge*: N parallel, **isolated** agent calls — no shared context — each receive the problem through one of **15 cognitive frames** deliberately distorted (with selection logic and custom frames), under a system prompt that **forbids evaluating**. *Focus*: a **separate** critic, with an opposite system prompt, scores the ideas (originality, viability, fit), clusters them by underlying angle, **flags the traps with their reasons**, and deepens the best survivors. The generator/critic separation is "mechanical": distinct LLM calls, not roles simulated within a single context — the same intuition as the separate-context adversarial review of the Bun project ([[sumner-bun-rewrite-rust-claude-2026-07-08]]).

The signature demo compares, on "a CLI that calls an LLM and sometimes freezes for 90s," the baseline (4 textbook patterns: progressive timeouts, exponential backoff, hedged requests, streaming — "the answer a senior gives in 30 seconds") against ADHD: 30+ ideas across 6 clusters, **20 named traps**, and a non-obvious pick — the **"rage-quit"** button that animates with the wait and instantly reroutes the request to a faster, cheaper model, because "the slow model might just be the wrong model for this prompt." On 6 open problems, the author's evaluation (LLM judge) gives breadth 9.00 vs 4.83, novelty 7.83 vs 2.67, **trap detection 9.50 vs 1.83**, actionability 9.50 vs 6.50 — self-reported figures, to be read as claims.

Distribution runs through the **skills** ecosystem (same channel as [[skill-pocock-grill-with-docs-2026-06]]): `npx skills add UditAkhourii/adhd` auto-detects ~50 agents (Claude Code, Cursor, Codex, Cline, Gemini CLI, Windsurf…), invocation via `/adhd` or auto-triggering on ideation intents, a CLI and an npm library, all built on the Claude and Codex Agent SDKs. Traction is tangible: a feature on The New Stack, a preprint, adoption by repowire (PR #313 merged — the frames become "peers" of the mesh-orchestrator), mstack (`think` plugin), zk-flow-oss, and an independent research review (testdouble/han) whose findings live on in public issues. Takeaway: useful divergence isn't prompted, it's **architected** — through context isolation and the mechanical generator/critic opposition.

## GrapheDeConnaissance

- Udit Akhouri —a_créé→ ADHD (TECHNOLOGIE, 0.98)
- ADHD —est_instance_de→ skills (TECHNOLOGIE, 0.95)
- ADHD —résout→ convergence prématurée (CONCEPT, 0.9)
- ADHD —utilise→ idéation divergente parallèle (METHODOLOGIE, 0.95)
- ADHD —est_variante_de→ Tree-of-Thought (METHODOLOGIE, 0.85)
- ADHD —utilise→ séparation générateur-critique (CONCEPT, 0.95)
- ADHD —utilise→ Claude Agent SDK (TECHNOLOGIE, 0.9)
- ADHD —s_applique_à→ Claude Code (TECHNOLOGIE, 0.95)
- ADHD —s_applique_à→ Cursor (TECHNOLOGIE, 0.9)
- convergence prématurée —observé_dans→ Chain-of-Thought (METHODOLOGIE, 0.9)
- convergence prématurée —observé_dans→ Tree-of-Thought (METHODOLOGIE, 0.9)
- ADHD —recommande→ "isolation totale des contextes pendant la divergence — les branches ne se voient jamais, donc pas d'ancrage" (AFFIRMATION, 0.95)
- ADHD —mesure→ "breadth 9,00 vs 4,83 ; novelty 7,83 vs 2,67 ; trap detection 9,50 vs 1,83 ; actionability 9,50 vs 6,50 — 5 victoires sur 6 problèmes ouverts (juge LLM, éval de l'auteur)" (MESURE, 0.7)
- repowire —utilise→ ADHD (TECHNOLOGIE, 0.95)

---
Canonical: https://www.thekb.eu/en/fiches/akhouri-adhd-ideation-divergente-parallele-2026-07-20/