researchlisted

# Research You investigate an open question and produce a **cited research report**. Research is read-only: you gather and synthesize evidence, you do not change code. > **Harness note:** Some harnesses (e.g. GitHub Copilot CLI) provide `/research` as a built-in agent — there this Skill is not installed. On harnesses without a built-in (e.g. Claude Code), this Skill provides the capability. ## Execution **Delegate** the investigation to a Sub-Agent with a **read-only toolset** (search, fetch, read — no edit, no build). The Sub-Agent has a fresh context window and runs the investigate→synthesize loop. Show the user the complete report. **Timeout:** Research is either fast or stuck. If the Sub-Agent shows no progress after 5 minutes, abort and do it yourself (Rule 26 / Rule 8). ## Input A **research prompt** — ideally the copy-paste-ready prompt generated by `/analyze` (persisted in the session, e.g. `files/research-prompt.md`). If no prompt is given, ask the user what exactly should be researched, then write the prompt yourself before starting. ## Sources — investigate three in parallel 1. **Codebase** — search the local project for existing implementations, patterns, libraries, and formats BEFORE declaring anything "new ground". Tools: Grep, Glob, Read. 2. **Web** — official documentation, framework guides, community patterns, real-world production experience. Tools: the harness's web search and fetch capability (e.g. WebSearch + WebFetch). 3. **Public repositories**