lucky-research

Build a single self-contained HTML dashboard of all captured ideas and every paper's pipeline progress — per-section completion, conference name, deadline, and days-left, with click-to-sort columns. Use when the user says "show my dashboard", "overview of my ideas and papers", "which paper is due first", or triggers /dashboard.

Design and run experiments bound to a GitHub repo. Tracks the bound repo's URL + SHA, records versioned execution attempts (semver), and mirrors structured results locally so /paper can consume them when writing. Use when the user wants to design, log, sync, or analyze experiments. Triggered by /experiment.

Generate research figures — structural SVG (architecture / pipeline / concept) or data matplotlib plots — plus curate a reference-figure library. Each figure goes through a 6-step interactive flow (intent → kind → refs → size → palette → render) and lands in the correct paper or experiment scope. Triggered by /figure.

Socratic research-direction tool — walk the user through Socratic discussion, scout the last 3 years of arXiv + WebSearch, score value/feasibility, recommend venues, emit a brain-library index, hand off to /paper. Each idea is persisted as an on-disk manifest under outputs/idea-checks/<slug>/ and mirrored to AgentDB ideas/<slug>. Also retains the legacy horizontal-matrix and vertical-lineage modes. Use when the user says "help me find a research direction", "is this idea novel", "compare these N papers", "what should I learn for this idea", "where does this idea come from", or triggers `/idea-check`.

Summarize research papers (PDF or arXiv) into structured markdown (problem / method / results / contribution / limitations) and index each summary into the AgentDB `papers/` namespace for later semantic search. Use when the user drops PDFs into `inputs/papers/` or pastes an arXiv URL.

Bundle per-user state into a zip for cross-machine migration and restore it on the new machine without overwriting existing files. Excludes externally-reproducible artifacts (HuggingFace base models, etc.) registered via per-experiment external-artifacts.md. Triggered by /migrate.

Multi-stage paper output flow — venue setup, direction discussion, paper scouting, problem focusing, motivation experiment design, section drafting. Organizes everything under outputs/papers/<venue>/<direction>/. Use when the user is starting or working on a paper for a specific venue. Triggered by /paper.

Manage references (BibTeX merge, cite-as-you-write resolution) and convert documents between Markdown / LaTeX / docx via pandoc. Use when the user says "build my bib file", "resolve [@cite] placeholders", "convert this draft to LaTeX / Word".

Long-running research mentor + boss-prep (科研导师 / 发展规划 / 老板汇报). Tracks self-trajectory (stated goals, weekly check-ins, decisions, drift) AND boss context (profile, meeting log, rehearsal) in AgentDB `project/` namespace. Use when the user says "check in", "what should I be working on this week", "am I on track", "回顾本周进度", or "/mentor boss …" / "/boss …" for boss-prep flows.

Generate publication-grade LaTeX algorithm pseudocode (algorithm + algpseudocode, or algorithm2e) for the current /paper or /experiment scope. Each algorithm goes through a 5-step interactive flow (intent → kind → I/O → draft → lint+render) and lands in <scope>/algorithms/<slug>.{tex,note.md,pdf}. Triggered by /pseudocode. Notation and venue conventions imported from https://github.com/HuiyuLi-2000/gen-pseudocode-skill (MIT).

OPTIONAL ruflo accelerator — parallelize paper scouting with a ruflo swarm of researcher agents, writing the SAME artifacts as `/paper scout` (related-papers/<slug>.md + AgentDB papers/<slug>). Use ONLY when ruflo swarm tools are available; degrades gracefully to suggesting `/paper scout` when they are not. Triggered by /scout-swarm.