minimal-adaptation

Implements the smallest literature-grounded code change for one selected research candidate.

Updated Aug 18, 2026
One-click install
npx skills add https://github.com/Nitrogen216/awesome-dsh-mods --skill minimal-adaptation-nitrogen216
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: minimal-adaptation
Source: https://github.com/Nitrogen216/awesome-dsh-mods/tree/main/modes/dsh-autoresearch/skills/minimal-adaptation
Command: npx skills add https://github.com/Nitrogen216/awesome-dsh-mods --skill minimal-adaptation-nitrogen216

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Research projects often stall or become unattributable when too many changes are made at once. This Skill enforces a disciplined adaptation phase where exactly one mapped candidate from the literature is implemented as the smallest possible, attributable modification before any screening training begins. ## Core Features & Use Cases - Structured Adaptation Spec: Writes a research/ADAPTATION.md file capturing the candidate id, source papers, causal path to the primary metric, files to change, and rollback path. - Narrow Implementation Guardrails: Restricts changes to one coherent primitive, preserves the fixed evaluation protocol and protected data split, and forbids tuning multiple alternatives at once. - Phase-Gate Workflow: Advances adaptation -> screening only when the candidate is runnable and attributable, or returns to mapping with outcome invalid if the mechanism cannot be implemented as mapped. - Use Case: After contribution mapping identifies a promising attention mechanism from a paper, use this Skill to implement just that mechanism in the project codebase, add focused unit tests, run a smoke test, and record the diff before any training. ## Quick Start Ask the agent to select one mapped candidate and implement its smallest attributable adaptation for screening.

Frequently Asked Questions about minimal-adaptation

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a paper's method in an existing research codebase?▼

Start by writing an ADAPTATION.md spec that records the candidate id, source papers, exact project-specific modification, and expected causal path to the primary metric. Then make the smallest source, config, and documentation change that realizes the mapped hypothesis, and add focused unit tests for the new behavior.

How to keep a research code change attributable to one hypothesis?▼

Limit the adaptation to one coherent primitive and never attach unrelated rescue modules or tune multiple alternatives in one implementation. Preserve the fixed evaluation protocol and protected split, and record the implementation diff plus any deviation from the paper.

When should I run full training after adapting a research candidate?▼

Do not launch full training during the adaptation phase. Only run a smoke test that exercises the changed path, then advance to the screening phase with outcome implemented once the candidate is runnable and attributable.

What happens if the mapped mechanism cannot be implemented?▼

Return to the mapping phase with outcome invalid rather than forcing a workaround. If the underlying assumption itself was wrong, refresh the literature before re-mapping the candidate.

Can I use an external oracle for implementation details during adaptation?▼

No. The adaptation phase forbids calling Oracle for code structure, interfaces, debugging, or implementation details. Major-decision escalation belongs to the mapping phase through the askgpt-governor.