research-question

# Skill: Research Question Formulation ## When Not to Use - When a research question already exists and has been validated — proceed directly to the `research` skill - When the task is to investigate rather than to scope (this skill prepares the question; `research` executes it) - When the question is a creative brief, design prompt, or opinion request — those are not research questions --- ## Interaction Protocol **Before starting**, ask if not already clear: 1. What decision or problem does the answer need to inform? 2. Are there known constraints (time, source types, access, domain)? 3. What output type is expected (knowledge, decision, recommendation, artefact)? **Output style**: Return a validated research question with scope, constraints, approach, and a readiness verdict. Flag every gap that would prevent investigation from starting. --- ## Inputs and Outputs **Input**: A candidate research question or topic statement, with optional context **Output**: A validated research question with scope boundaries, constraints, decomposed approach sub-questions, and a readiness verdict (ready / needs revision) **Composability**: Use before `research` (gates the start of investigation); the output feeds directly into `research` §0 Initialise --- ## 1. Question Quality Test A research question is ready when it passes all five tests: | Test | Pass condition | Fail signal | |---|---|---| | **Specific** | Names the subject, the comparison or evaluation, and the context | "Investigate X", "What about Y", vague topic statements | | **Answerable** | Can be resolved with evidence available in principle | Requires future events, unobtainable data, or pure opinion | | **Scoped** | Has explicit in-scope and out-of-scope boundaries | No scope defined; would expand infinitely | | **Motivated** | States what decision or problem the answer informs | No stated purpose; investigating for its own sake | | **Decomposable** | Can be broken into at least two atomic sub-questions | Monolithic; cannot be subdivided | If any test fails, the question needs revision before investigation begins. --- ## 2. Question Rewriting If the question fails the Specific test, rewrite it using this pattern: > "What is the [best / most appropriate / correct] [X] for [purpose] given [constraints]?" Examples: - Weak: "Investigate memory management for agents" - Strong: "What is the most operationally practical memory architecture for a single-owner research agent running in GitHub Actions, given a constraint of no persistent server and a corpus under 500 items?" If the question fails the Scoped test, add explicit boundaries: - **In scope**: list the specific aspects, technologies, or contexts that are in scope - **Out of scope**: list the aspects explicitly excluded and why - **Constraints**: time horizon, source types, access limitations, budget --- ## 3. Approach Decomposition Decompose the validated question into sub-questions before investigation begins. Rules: - Each sub-question must be answerable with a single evidence-based claim - Sub-questions must be independent where possible (parallel, not sequential) - Number each sub-question; use letter suffixes for nested decomposition (1a, 1b) - If a sub-question cannot be answered without first answering another, make the dependency explicit Stopping condition: decomposition is complete when every leaf sub-question passes the Specific and Answerable tests independently. --- ## 4. Readiness Verdict After applying tests, rewriting, and decomposing, issue one of two verdicts: **READY**: The question is specific, answerable, scoped, motivated, and decomposed. Hand to the `research` skill. **NEEDS REVISION**: List each failing test with a concrete fix instruction. Do not proceed to investigation until all tests pass. --- ## 5. Output Format ``` Research Question: [final validated question] Scope: In scope: [list] Out of scope: [list] Constraints: [list] Context: [one sentence: what decision or problem this informs] Approach: 1. [sub-question] 1a. [atomic] 1b. [atomic] 2. [sub-question] Readiness: READY | NEEDS REVISION [If NEEDS REVISION: list each failing test and fix instruction] ```