decompose-constrained-set-search
$
npx mdskill add Memento-Teams/Memento-Widesearch/decompose-constrained-set-searchFinds a specific entity by filtering through multiple independent constraints in a multi-stage verification process.
- Helps locate a single item meeting distinct attributes like timeframe, thresholds, exclusions, and affiliations simultaneously.
- Integrates with or depends on unspecified tools for candidate generation and fact-checking passes.
- Decides by extracting constraints, prioritizing the most restrictive filter, and sequentially applying others with deep verification.
- Presents results through a final candidate list after thorough auditing of all original requirements.
SKILL.md
.github/skills/decompose-constrained-set-searchView on GitHub ↗
--- name: decompose-constrained-set-search description: Orchestrates a multi-stage filter-and-verify process to find a specific entity satisfying three or more independent constraints. --- ## When to Use Use this skill when the user query defines a "needle in a haystack" search requiring the intersection of multiple distinct attributes. Indicators include queries for a single entity (person, place, object, or document) that must satisfy a specific timeframe, a quantitative threshold, a negative constraint (exclusion), and a specific affiliation or category simultaneously. ## Decomposition Template 1. **Constraint Extraction**: Parse the query into a checklist of independent constraints (e.g., Constraint A: Timeframe, Constraint B: Quantitative Metric, Constraint C: Exclusion/Negative, Constraint D: Affiliation). 2. **Primary Candidate Generation**: Identify the "most restrictive" or "easiest to index" constraint. Generate a broad list of candidates that satisfy this single primary filter. 3. **Sequential Filtering**: Pass the candidate list through the remaining constraints one by one. 4. **Deep Verification**: For the final 1-2 candidates, perform a dedicated "fact-check" pass against every original constraint, specifically looking for "hidden" requirements or negative exclusions. ## Worker Assignment Rules - **Worker 1 (The Scout)**: Responsible for Step 1 and 2. Must prioritize recall to ensure the "needle" is in the initial set. - **Worker 2 (The Auditor)**: Responsible for Step 3. Cross-references the candidate list against secondary data points. - **Worker 3 (The Critic)**: Responsible for Step 4. Acts as a "Red Team" to try and disqualify the final candidate based on the most easily overlooked constraint (e.g., the negative constraint). ## Answer Format The final output must clearly state how the chosen entity satisfies every constraint listed in the prompt. End with: `最终答案:你的答案` ## Anti-Patterns - **The "Close Enough" Fallacy**: Selecting a candidate that meets 3 out of 4 criteria because it is a high-probability match in other contexts. - **Primary Filter Failure**: Starting the search with the broadest constraint (e.g., "a person") rather than the most specific one (e.g., "published in October 2023"), leading to an unmanageable candidate pool. - **Negative Constraint Oversight**: Forgetting to check "Except X" or "Not including Y" during the final verification phase. - **Hallucinated Aggregation**: Combining attributes from two different entities to create a "perfect" but non-existent match.