xd-liu

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 1 of the research-idea pipeline. Expands a seed meta-idea into 10 distinct idea variants and a set of literature-search queries. Spawned by the research-ideas orchestrator.

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 6 of the research-idea pipeline. Adversarially reviews each generated paper idea — strengths/novelty/impact, defects, and overlap with the retrieved literature — and assigns credit/blame to the pipeline step that produced each. Spawned (often in parallel) by the research-ideas orchestrator.

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 4 of the research-idea pipeline — the 杂交衍生 core. Cross-breeds paper "genes" (and the brainstorm variants) into novel candidate paper ideas with explicit key insights. Spawned (often several in parallel) by the research-ideas orchestrator.

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 5 of the research-idea pipeline. Scores all candidate ideas on novelty/feasibility/impact, dedupes near-duplicates, and ranks them by priority. Spawned by the research-ideas orchestrator.

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 3 of the research-idea pipeline. Reads the retrieved papers and digests each into a reusable structured "gene". Runs in-context on Sonnet (your Claude Code subscription) — no API key needed. Spawned by the research-ideas orchestrator.

xd-liu/AutoResearch-Idea

Agent Claude Code

Step 2 of the research-idea pipeline. Runs the search tool to retrieve, dedupe, and rank 100 related papers from the brainstorm queries. Spawned by the research-ideas orchestrator. Cheap model — it just drives a tool.

xd-liu/AutoResearch-Idea

Agent Claude Code

Meta/development logging agent — NOT part of the research-idea pipeline. Appends a concise entry to IMPROVEMENTLOG.md capturing the session's user prompts, key insights, and issues+fixes, so the system can be improved later. Invoke manually after notable work (e.g. "use the self-improvement agent to log this session").

xd-liu/AutoResearch-Idea

Hook Claude Code

Runs when you submit a prompt, before the agent sees it, running an inline shell check (2 commands). From xd-liu/AutoResearch-Idea.

xd-liu/AutoResearch-Idea

Settings file Claude Code

Agent settings declaring 1 hook event (UserPromptSubmit).

xd-liu/AutoResearch-Idea

Skill Claude CodeCodex

Orchestrate the research-idea pipeline: turn a seed meta-idea into a ranked list of 50-100 novel paper ideas. You delegate each step to a dedicated subagent (brainstorm → retrieve → digest → hybridize → prioritize), tracking progress in a run directory that the live web dashboard reads. Use this whenever the user…

xd-liu/AutoResearch-Idea

Instructions file

Claude Code instructions for xd-liu/AutoResearch-Idea, covering claude.md, what this is, two constraints that will bite you if ignored, architecture (the big picture) and commands.