What Are the 15 Cognitive Frames in ADHD and How Does Frame Selection Work?

The ADHD repository defines 15 built-in cognitive frames—deliberate "vantage-operator" prompts that re-pose problems from different conceptual angles—and selects them via a deterministic, bias-aware algorithm in src/frames.ts that always includes at least one "wild" perspective.

The open-source UditAkhourii/adhd project implements a parallel reasoning engine that uses cognitive frames to force LLM generators into unconventional conceptual corners. These frames are small system-prompt payloads defined in the source code that deliberately shift perspective, from "Hardware Engineer" to "10-Year-Old," to maximize divergent thinking. Understanding the 15 cognitive frames in ADHD and their selection mechanism is essential for customizing the engine's creative output.

The 15 Built-In Cognitive Frames in ADHD

The ADHD project maintains a curated set of frames in src/frames.ts, each designed to break fixation by imposing a foreign mental model. Every frame carries a unique ID, descriptive label, vantage prompt (≈5 lines), and categorization tags.

Frame ID	Label	Vantage (Prompt Summary)	Tags
hardware-eyes	Hardware engineer	Think in latency, memory layout, physical constraints.	code, wild
regulator	Regulator / auditor	Audit for provable, traceable, refusable aspects.	design, general
ten-year-old	10‑year‑old	Naïve, unencumbered approach; ignore conventions.	general, wild
adversary	Competitor trying to break it	Generate adversarial exploits and then invert them.	code, design
biology	Cross‑domain: biology	Transplant mechanisms from immune systems, neural plasticity, etc.	code, wild
logistics	Cross‑domain: logistics / supply chain	Apply queues, batching, just‑in‑time, hub‑and‑spoke ideas.	code, design
game-design	Cross‑domain: game design	Identify loops, rewards, friction, speed‑run tricks.	design, general
markets	Cross‑domain: markets	Model the problem as an auction, futures contract, clearing house.	design, wild
inversion	Inversion	Ask the opposite question, then negate the answers.	code, design, general
extreme-zero	Extreme: $0 budget, 1 hour	Produce the crudest workable version under severe constraints.	code, general
extreme-infinite	Extreme: infinite budget, 10 years	Imagine a maximalist solution with unlimited resources.	design, wild
remove-assumption	Remove the load‑bearing assumption	Drop the assumed infrastructure (framework, DB, network) and redesign.	code, design, wild
speedrunner	Speedrunner	Find glitches, skips, frame‑perfect shortcuts.	code, wild
ant-colony	Ant colony / swarm	Use decentralized local rules and emergent behavior.	code, wild
ops-3am	On‑call at 3 am	Design to avoid paging; create a runbook‑friendly solution.	code, design

Full descriptions and prompt payloads are documented in documentation/frames.md according to the repository source.

How Frame Selection Works

The selection algorithm lives in the selectFrames function within src/frames.ts. The process combines deterministic randomness with deliberate bias to ensure both reproducibility and creative divergence.

Code-Mode Bias and Tag Filtering

When codeMode is true (the default), the algorithm restricts the primary candidate pool to frames carrying the code or design tags. This prioritizes engineering-centric perspectives. Disabling codeMode expands the pool to include all 15 frames regardless of tag.

Wildcard Reservation for Divergent Thinking

Frames tagged with wild are always reserved in a separate pool. The algorithm guarantees that at least one wild frame appears in every selection to ensure "weird," non-engineered perspectives surface during generation.

Deterministic Shuffling Algorithm

The selectFrames implementation follows these exact steps:

1. Filter by mode – Apply the codeMode tag filter to create the primary pool.
2. Separate wildcards – Isolate all wild tagged frames into a reserved pool.
3. Fisher-Yates shuffle – Shuffle the primary pool using a seedable random generator for reproducibility.
4. Pick N‑1 frames – Select the first n‑1 frames from the shuffled primary pool (or at least one if n is 1).
5. Insert wild frame – Draw a random wild frame and append it if not already present.
6. Trim to N – Slice the combined list to exactly n items.

The function signature is:

export function selectFrames(n: number, codeMode = true): Frame[] { … }

Key properties include deterministic per-seed output, guaranteeing identical frame sets across runs with the same seed, and the mandatory wild frame requirement, ensuring creative divergence even in code-heavy sessions.

Implementation in src/frames.ts

The frame definitions and selection logic reside in src/frames.ts, which exports both the FRAMES array and the selectFrames utility. The Frame type (implied by the implementation) includes id, label, vantage (the prompt text), and tags properties. The src/engine.ts file orchestrates parallel LLM calls using these selected frames, invoking selectFrames at runtime to populate the vantage prompts.

Practical Usage Examples

Below are runnable TypeScript snippets demonstrating how to retrieve frames and invoke the selection logic.

Retrieve the full list of available frames:

// Example 1 – Get the full list of frames
import { FRAMES } from "./src/frames";
console.log("All frames:", FRAMES.map(f => f.label));

Select 5 frames with the default code-mode bias:

// Example 2 – Select 5 frames with the default code‑mode bias
import { selectFrames } from "./src/frames";

const chosen = selectFrames(5);          // deterministic per seed
console.log("Chosen frames:");
chosen.forEach(f => console.log(`- ${f.label} (${f.id})`));

Disable code-mode to expose every frame type:

// Example 3 – Disable code‑mode to expose every frame
const allKinds = selectFrames(4, false);
console.log("All‑kind selection:", allKinds.map(f => f.id));

These snippets can be executed via the repository’s CLI using npm run ts-node or integrated into larger orchestration pipelines.

Summary

• The UditAkhourii/adhd repository defines 15 cognitive frames in src/frames.ts, ranging from "Hardware Engineer" to "Ant Colony," each imposing a unique conceptual vantage.
• Frame selection occurs through the selectFrames function, which filters by tags, applies a Fisher-Yates shuffle, and guarantees at least one wild tagged frame.
• Code-mode bias (codeMode = true) restricts selection to engineering-relevant frames by default, while disabling it exposes the full spectrum.
• The algorithm is deterministic per seed, enabling reproducible experiments with identical frame sets across runs.
• Selected frames feed into src/engine.ts to drive parallel LLM reasoning with deliberately divergent perspectives.

Frequently Asked Questions

What is the purpose of cognitive frames in ADHD?

Cognitive frames serve as deliberate "vantage-operator" prompts that re-pose a problem from a different conceptual angle. Each frame is a small system-prompt payload (≈5 lines) that forces the generator into a corner it would not naturally explore, such as thinking like a hardware engineer or a speedrunner.

How does the frame selection algorithm ensure diversity?

The algorithm ensures diversity by reserving slots for frames tagged with wild, which represent unconventional or cross-domain perspectives. It also uses a Fisher-Yates shuffle for randomization while maintaining deterministic output based on a seed, guaranteeing that every run includes at least one non-engineered, creative viewpoint.

Can I add custom frames to the ADHD repository?

Yes, the modular structure of src/frames.ts allows extending the exported FRAMES array with custom Frame objects. Each custom frame must include an id, label, vantage prompt string, and tags array following the existing type definitions used by the selectFrames function.

What is the difference between code mode and general mode in frame selection?

Code mode (the default, codeMode = true) filters the candidate pool to frames tagged with code or design, prioritizing practical engineering and implementation perspectives. General mode (codeMode = false) removes this filter, exposing all 15 frames including those tagged only with general or wild, resulting in broader conceptual diversity.

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