explore
Adversarially explore a request before formalizing work. Retrieve repository, memory, and external prior art; use Context7 for explicitly required technologies; expose conflicts, edge cases, user acti
SKILL.mdskills/explore/SKILL.md
GitHubExplore
Turn an idea or problem into a trustworthy recommendation. Do not implement or create a formal change until the request is coherent enough to specify.
Stages
- Frame intent, outcome, constraints, and explicitly required technologies.
- Retrieve internal prior art from active and archived changes, repository code/docs/tests, and relevant project memories. Treat memories as context and revalidate them against current state.
- When a new library, framework, language, API, or specific function is explicitly required, consult Context7 for version-relevant documentation. Use broader web research when Context7 lacks coverage or ecosystem evidence is needed.
- Inspect current callers, consumers, contracts, integrations, configuration, and tests.
- Run an adversarial pass: try to disprove the approach by checking conflicts, edge cases, security, permissions, failure states, concurrency, migrations, compatibility, accessibility, performance, observability, and testability.
- Identify user-only actions, external approvals, credentials, provisioning, irreversible choices, and unresolved decisions. Resolve them now or split dependent work.
- Compare alternatives and recommend one with explicit trade-offs.
- Produce a durable exploration record and route to feature authoring or stop with the recommendation.
Output
Record intent, internal prior art, memory context, Context7 findings, external evidence, current state, adversarial findings, user actions/decisions, alternatives, recommendation, open questions, proposed scope, and verification concerns.
Never treat an old memory, search result, or worker claim as proof without current evidence.