techdufus Ultraresearch
User message: research request. Execute as researcher under this contract; it overrides stopping at the first hit or a single source.
Ultraresearch
Exhausts sources with parallel WebSearch, WebFetch, and explore, then cross-checks and cites. Use when the user invokes /ultraresearch or needs thorough external plus repository research before deciding.
Ultraresearch by techdufus · 541981c
npx skillmds@latest add techdufus/ultraresearch File contents
---name: ultraresearchdescription: Exhausts sources with parallel WebSearch, WebFetch, and explore, then cross-checks and cites. Use when the user invokes /ultraresearch or needs thorough external plus repository research before deciding.---# UltraresearchUser message: research request. Execute as researcher under this contract; it overrides stopping at the first hit or a single source.<role>Exhaust independent sources, then synthesize. Prefer parallel `WebSearch`/`WebFetch` in one turn plus `explore` for repository evidence. Cite what you used. Do not implement unless the user also asked for a change.</role><rules>1. Enumerate the questions to answer before searching. "Look into X" is not a search plan.2. Fan out: multiple `WebSearch` queries and `explore` in one message. Follow with `WebFetch` on the best hits. NEVER stop at one snippet.3. Cross-check. A claim that appears in one page is provisional until a second independent source or repository evidence agrees.4. Separate repo facts from web claims. Repository evidence wins for *this* codebase; web evidence wins for upstream docs, RFCs, and ecosystem practice.5. Cite: URL or file:line for every material claim. Do not invent sources.6. Record unresolved disagreements instead of averaging them away.7. Stop when questions are answered or `[blocked]` on missing access. Do not pad with adjacent topics.8. Right-size: a single known-doc lookup is `WebFetch` or `Read`, not a research program.</rules><workflow>1. Scope: list the questions and what would falsify each.2. Sweep: parallel web + `explore`.3. Deep-read: `WebFetch` / `Read` the hits that matter.4. Synthesize: answers, citations, conflicts, unknowns.5. Yield terse findings, not a recap of every page.</workflow><anti-patterns>- One `WebSearch` then answering from the snippet.- Treating search-result titles as verified facts.- Mixing this skill into an implementation fan-out; research first, then implement via Task.- Dumping uncited summaries.</anti-patterns>
techdufus/dotfiles/tree/main/roles/cursor/files/skills/ultraresearch commit 541981cd96
Frequently asked questions
Run npx skillmds@latest add techdufus/ultraresearch in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Exhausts sources with parallel WebSearch, WebFetch, and explore, then cross-checks and cites. Use when the user invokes /ultraresearch or needs thorough external plus repository research before deciding. It is listed under Research & Search on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
techdufus (@techdufus) published this skill. Their other Agent Skills are listed on their SkillMD profile.