long-research
от vanya1210
Web & Frontend DevelopmentКратко
[BETA] Deep research that actually reads pages instead of summarizing search results.
Описание
Что делает этот навык
Deep research that reads full pages rather than summarizing search snippets. A tree-style loop spawns new searches from each page read, and a hard read-to-search ratio prevents shallow search-spamming. A self-audit gate blocks delivery until quality checks pass.
It enforces actual page reads rather than relying on AI-synthesized search summaries, so findings trace back to fetched content with verifiable sources.
Когда его использовать
- Vetting a supplement with conflicting forum reports before buying
- Finding firsthand accounts of a rare visa denial or approval situation
- Comparing real user experiences with competing software tools before switching
- Pulling current prices from retailer sites for a high-value purchase decision
- Digging into login-gated forums for niche hardware troubleshooting
Пример сценария
User asks: Research mechanical keyboards under $150 — find real forum opinions, not marketing copy. Spend 20 minutes.
- 1
Confirms topic, duration, mode, tools, and output format via pre-flight checklist
- 2
Posts a research plan block and waits for explicit approval before making any tool calls
- 3
Runs survey phase to identify at least 10 keyboard candidates across forums and review sites
- 4
Reads forum threads via web_fetch, escalating to browser-use cascade for blocked or JS-heavy pages
- 5
Runs self-audit gate to verify read/search balance, source quality, and task fidelity before delivering
Research file at research/mechanical-keyboards-[date].md with verified findings, direct forum quotes, and a comparison table; chat summary under 500 words linking to the file
Требования
- browser-use installed via: pip install browser-use && browser-use install
Установка
clawhub install vanya1210/long-researchВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.