goal-driven-decomposition

Goal-Driven Decomposition

Citation convention used in this skill Inline references look like [Author Year, §section] or [Source]. They sit next to the specific claim they support, not at the end. The full reference list is in the References section at the bottom. Claims without a citation are either common knowledge or distilled from this skill's own iteration history (a real conversation in which an LLM repeatedly over-engineered a "Monte Carlo simulation website" request before being corrected — that conversation is the seed case for this skill, recorded in the Appendix).

What this skill exists for

LLMs systematically over-engineer when given short goals. Given "build a Monte Carlo simulation website" the default failure mode is to immediately produce a Kubernetes-grade architecture (Redis Streams, worker pools, Rust modules) instead of recognizing that "website" + "simulation" + casual phrasing implies a teaching demo that should run entirely in the browser. (This is the seed case — see the Appendix at the end for the full failure trace.)

This is a known failure pattern in the literature, called premature decomposition or overengineering:

• Decomposition has a hidden coordination cost. Amazon Science formalized this as O(n) + O(k^m) where k is the number of subtasks and 1 < m ≤ 2. For small k the overhead is negligible; as k grows, coordination cost dominates and destroys the gains from decomposition. They also note that excessive decomposition causes the system to "fail to capture the serendipitous connections and novel insights that can emerge from a more holistic approach." [Gozluklu 2024, Amazon Science]
• Eager planning fails when subtasks turn out to be unexecutable. Plan-and-execute approaches commit to a full decomposition up front; if any subtask fails, the whole plan fails. [Prasad et al. 2024, ADaPT §1]
• The fix is to decompose only on failure. ADaPT "explicitly plans and decomposes complex sub-tasks as-needed, i.e., when the LLM is unable to execute them" — outperforming ReAct and Plan-and-Solve by up to 28–33 percentage points on ALFWorld, WebShop, and TextCraft. [Prasad et al. 2024, ADaPT §5]