research · git:20261102.bc6f8af · 2026-11-02 · sha256 4071156df5e97d9c
research git:20261102.bc6f8afA
Immutable. This exact content never changes and is served at /api/v1/blob/4071156df5e97d9c.
No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_o3bxworhcxuedcdm/4071156df5e97d9c17bcde7c0cf1d6da9e766a952e3d4416e03e8ce10471df29.json
Then /plugin install research@markdownregistry-o3bxworhcxuedcdm in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.
--- name: research description: Investigates technical questions, library APIs, and modern architecture against high-trust primary sources via live search and deep URL ingestion. Use when researching documentation, best practices, latest framework changes, or external system integrations. --- # Technical Research & Empirical Grounding Investigate complex technical topics, modern library conventions, architecture patterns, and official documentation against high-trust primary sources using live web search and deep URL ingestion. ## Decision tree 1. **Does the task require live rendering of a browser DOM, visual UI layouts, or interactive clicks?** - **Yes** → Use `/browser` or the `browsing` skill. - **No** → Proceed to step 2. 2. **Is the research question strictly about local codebase symbols, files, or internal data flows?** - **Yes** → Delegate to `codebase-investigator` via `invoke_subagent` or inspect locally via `grep_search` and `read_file`. - **No (external frameworks, 2026 best practices, third-party libraries, modern APIs, or platform specs)** → Execute the 4-phase research workflow below. ## Research workflow ### Phase 1: Question decomposition & year-anchored query matrix - Deconstruct the user's inquiry into 2–3 specific, orthogonal search queries. - Anchor queries with the explicit year (`2026`), latest version numbers, and authoritative domains (e.g. `site:github.com`, `site:docs.<framework>.org`, official changelogs). - Avoid vague keywords; target RFCs, official API references, and primary migration guides. ### Phase 2: Live web search & exhaustive primary ingestion - Execute search queries via `search_web`. - **Anti-Snippet Skimming Invariant**: Never guess or synthesize findings from brief 50-word search snippets. - Ingest full primary documentation pages using `read_url_content` (or fetch via `run_command` if needed) to examine exact method signatures, types, and constraints end-to-end. ### Phase 3: Cross-source triangulation & falsification - Cross-reference claims across at least 2 independent authoritative sources (e.g. official docs + release changelog + GitHub discussion/issue). - Eliminate deprecated 2022–2024 patterns (e.g. legacy Next.js Pages router, outdated React hooks, obsolete CLI flags). - Verify licensing, dependency footprint, and cross-platform compatibility (Linux, macOS, Windows). ### Phase 4: Structured evidence synthesis & artifact generation - Synthesize findings into a structured report containing: - **Executive Verdict & Recommended Pattern**: The single best-practice architecture adhering to the 6 Engineering Pillars. - **Empirical Evidence & Primary Sources**: Verifiable URLs with dates and line/section citations. - **Concrete Drop-In Example**: Minimal working code or configuration. - **Deprecated Traps to Avoid**: Legacy patterns commonly hallucinated from 2024 training data. - Save complex multi-page research to `.scratch/research/<topic-slug>.md` as an observable artifact.