| name | research-question-framing |
| description | Use when a research topic has unclear scope, ambiguous terms, missing evidence criteria, or is likely to become a source-code learning study, investigation, audit, architecture explanation, or design inquiry |
Research Question Framing
Purpose
Turn a fuzzy research topic into an evidence-seeking brief before source-code study begins.
Core principle: Do not start reading broadly until the question, context, evidence standard, and test cases are explicit.
When to Use
Use before a source-code learning SOP, architecture investigation, performance diagnosis, agent-system study, or design research task when the user has a topic but not a clear research frame.
Do not use for single-symbol lookups, direct bug fixes, or tasks where the user already gave precise questions and evidence criteria.
Workflow
topic -> context -> divergent questions -> refined questions
-> evidence matrix -> validation scenarios -> handoff brief
- Frame context: user, decision, system, scope, non-goals, and why the answer matters.
- Check method basis: start with
references/best-practices.md; search externally only when requested, unfamiliar, high-stakes, unstable/current, or dependent on domain standards.
- Generate 8-12 candidate questions without answering them.
- Improve questions: turn vague yes/no questions into answerable open questions, and broad open questions into closed checks where needed.
- Prioritize 3-5 primary/subquestions plus 2-3 negative checks.
- Build an evidence matrix: source-code, runtime, docs, tests, traces, metrics, and negative evidence.
- Create 3-5 validation scenarios that pressure the topic's exact ambiguity.
- Handoff only after evidence criteria; likely entry points come last.
Minimal Response Mode
When the user asks for speed or says "just list files," still include:
- one-sentence topic,
- definitions for the ambiguous terms,
- 2-3 proof / non-proof criteria,
- likely entry points only after those criteria.
Blocked Source-Code Handoff
If a source-code study lacks the repo, branch/version, entrypoint, metric, or target behavior, do not hand off to source reading yet.
If the user says not to ask questions, state assumptions and missing inputs as requirements:
## Handoff Status
Blocked: missing {repo/version/entrypoint/metric/target behavior}.
## Required Inputs
- ...
## Assumptions If Forced To Proceed Later
- ...
Brief Template
# Research Brief
## Topic
One sentence.
## Situation
User/context, decision to support, constraints, and what confusion must be removed.
## Key Terms And Boundaries
Define ambiguous words and what would count as in/out of scope.
## Question Tree
- Primary question
- Subquestions
- Negative checks
## Evidence Matrix
| Question | Evidence needed | Source type | What would prove it | What would not prove it |
|---|---|---|---|---|
## Validation Scenarios
3-5 cases that should still produce a useful frame.
## Handoff To Source-Code SOP
Scope, likely entry points, required artifacts, and acceptance criteria.
Evidence Ladder
Use the strongest available evidence: runtime caller/state transition > config default/feature gate > integration/unit test > tool spec/API/command > prompt/instruction text > docs/README > absence after documented search.
Rules:
- Tool/API existence proves capability, not mandatory behavior.
- Prompt text proves guidance, not runtime enforcement.
- Source code alone rarely proves performance or production behavior; require traces, logs, metrics, or field data.
- Absence is only useful when search scope and terms are recorded.
Skill Deployment Test Cases
Use tests/scenario-bank.md before deploying or changing this skill. Validate at least:
- source-code agent workflow study,
- frontend first-load performance investigation,
- multi-agent spawn/context/merge study,
- ambiguous product/architecture research with no repo yet.
Red Flags
- "Read these files" appears before "what would prove the answer."
- The handoff cannot distinguish capability from mandatory behavior.
- The topic contains terms like "fixed workflow", "slow", "decide", "autonomous", or "architecture" without definitions.
- There is no
What would not prove it column.
- There are no test scenarios.
- The brief lacks repo/version/entrypoint assumptions for source-code studies.
- Performance claims do not separate lab, field, source, and runtime evidence.
- A bespoke process is invented before checking the local method reference.
Handoff Rule
Only hand off to a source-code SOP when the Research Brief has: context, boundaries, prioritized questions, evidence matrix, validation scenarios, and acceptance criteria.