Agent skill · ma08
deep-research
Run end-to-end deep research using OpenAI Responses API deep-research models plus Gemini Deep Research plus Exa Deep Research, then synthesize results into a comprehensive, easy-to-read, citation-backed Markdown report by default. Use for due diligence, investment research, market/competitor analysis, technical investigations, policy analysis, and any high-stakes question where cross-checking sources across multiple engines improves confidence; generate alternate formats only when the user explicitly requests them.
What it needs
About 6k tokens when loaded.
What this skill does
Deep Research Produce decision-grade research outputs with multi-engine evidence gathering: OpenAI deep research via bundled script (scripts/runopenaideepresearch.py) Gemini deep research via bundled script (scripts/rungeminideepresearch.py) Exa Research API via bundled script (scripts/runexaresearch.py) Default output is always a comprehensive Markdown report. If the user asks for additional formats, provide them after the Markdown report. Non-Negotiable Output Contract 1. Deliver a comprehensive Markdown report first. 2. Make the Markdown easy to scan: clear headings, concise tables, explicit assumptions. 3. Attach citations/links for all load-bearing claims. 4. Clearly separate evidence from inference. 5. Only add alternate outputs (JSON, CSV-style table, Notion-ready, slide-outline, short brief) when explicitly requested. Use references/output-contract.md for the exact structure. Speed Policy (Parallel-First) Optimize for shortest end-to-end time without reducing research quality. 1. Run OpenAI, Gemini, and Exa research in parallel whenever all are required. 2. Parallelize independent sub-questions with multiple agents (market, technical, legal, risks, benchmarks). 3. Start synthesis only after minimum evidence is available from all active engines, then continue ingesting late-arriving evidence in parallel. 4. Keep at least one high-level coordinator agent to merge claims, resolve conflicts, and enforce citation quality. 5. Use sequential execution only for truly dependent steps. Workflow 1. Scope and clarify Define audience, decision to support, time horizon, and constraints. Confirm required output formats. If unspecified, use Markdown only. 2. Prepare research brief Create a prompt file in the current task artifacts folder. Include required deliverables, confidence handling, and citation requirements. 3. Launch research engines in parallel OpenAI track: run scripts/runopenaideepresearch.py (prefer submitandcheck). …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills ma08/deep-research