BITE

Migrates BITE's architecture to a new professional domain (e.g. frontend development, accounting, journalism). Interactive session with the user to map research concepts to domain equivalents, then generates a complete set of adapted skills and folder structure. Explicit trigger only.

Batch-analyzes papers from a user-provided CSV or `obsidian-vault/paper_list.csv` by splitting the list into safe per-batch CSVs, spawning parallel worker agents, writing analysis notes under `obsidian-vault/analysis/`, and refreshing `obsidian-vault/index/` after all completed batches. Use when the user asks to analyze many papers from `paper_list.csv` or run a local batch paper analysis workflow.

Collects paper candidates from non-GitHub web URLs — conference sites, lab homepages, proceedings pages, Google Scholar results, blog posts with paper lists, etc. Fetches and stores pages locally, then appends rows to `obsidian-vault/paper_list.csv`. Use when the user provides web URLs (not GitHub repos) + keyword constraints + venue/year and wants candidates added to the unified log. For GitHub repos, use `papers-collect-from-github-repo` instead.

Uses colocated paper download tools to download, verify, and repair local PDFs from `obsidian-vault/paper_list.csv` rows so that `obsidian-vault/paperPDFs/` stays complete and deduplicated. Use when candidate rows are in `state=Wait` and should become `Downloaded` before the formal analysis chain.

Unified entry for the research pipeline. Maps current work to one stage (import-local-pdfs/collect/download/analyze/build/query/ideate/focus/review/audit/export), recommends the right existing skill/command, supports step-by-step and end-to-end guidance, and keeps stage boundaries clear without duplicating underlying capabilities.

Runs a unified metadata consistency audit across obsidian-vault/analysis logs and analysis notes, checking title/venue/year consistency, link/pdf_ref completeness, duplicates, and structure anomalies; writes quality_report_*.md under obsidian-vault/batches/reports.

Builds/refreshes `obsidian-vault/index/index.jsonl` and Obsidian navigation pages from `obsidian-vault/paper_list.csv` when present, enriched by `obsidian-vault/analysis/` frontmatter when available. Use after completed analysis batches or when users ask to rebuild the local paper index.

Obsidian Flavored Markdown conventions for BITE. Use when generating or editing any .md file in the vault (analysis notes, indexes, idea notes, wiki pages). Ensures all output is Obsidian-native and renders correctly in reading view, graph view, and backlink panels. Adapted from kepano/obsidian-skills (MIT).

Converts internal research notes into shareable Obsidian-friendly Markdown files. Use when the user explicitly asks to make a share version, export a note for sharing, hide knowledge base traces, replace internal wiki links with public paper links, rewrite internal wording for external readers, or generate a content-driven share filename.

Queries the local `obsidian-vault/analysis/` notes for research. Finds papers by title, task path, technique tags, venue, or year; summarizes methods and evidence across papers; cites core_operator and primary_logic from analysis frontmatter. When present, `obsidian-vault/index/index.jsonl` serves as the fast filter layer and `obsidian-vault/index/` pages serve navigation and backlink exploration.

Explores and structures research ideas into `obsidian-vault/ideas/` notes using the local paper knowledge base and frontier techniques. Use when the user provides research questions or ideas and wants decomposed candidate directions, related-work analysis based primarily on `obsidian-vault/analysis`, and cross-domain support from image/video generation, MLLM, Agent, or RL, automatically saved as dated idea notes under `obsidian-vault/ideas/`.

Retrieves code-context-relevant papers from the local knowledge base in two modes (brief/deep), using `obsidian-vault/analysis/` as the primary retrieval source and `obsidian-vault/paperPDFs/` for optional deeper reading. `obsidian-vault/index/` is optional and mainly serves statistics, Obsidian navigation, and backlink exploration. Environment detection prioritizes codebase environment files; asks user if none found. Trigger timing: BEFORE code modification, not after.

Collects paper candidates from any GitHub repository README or docs — awesome lists, survey companion repos, lab paper lists, conference accepted-paper repos, benchmark leaderboards, etc. Agent fetches the README (or specified doc), analyzes its format, writes a one-off parser, and outputs rows aligned with paper_list.csv. No pre-built scripts — agent adapts to each repo's format on the fly.

Generates or updates a structured daily research log by collecting evidence from git diffs, artifacts, and conversation context, then compressing into a decision-oriented summary. Use when the user says "写日志", "write daily log", "日报", "update daily summary", or asks to summarize today's research progress.

Guides research ideation from broad thoughts to a focused, testable plan with explicit scope cuts, prioritized hypotheses, and immediate next experiments. Use when the user wants co-creation and step-by-step narrowing, not review verdicts.

Runs a strict-but-fair ICLR/CVPR/SIGGRAPH reviewer challenge on idea, roadmap, or full paper, with major-risk diagnostics and actionable repair paths. Use when the user wants high-pressure review questioning instead of brainstorming.

Detects post-invocation skill mismatch patterns, summarizes why they may recur, and asks whether to revise the skill now. Use when a called .claude skill output is clearly unfit for the user intent and similar mismatch is likely to happen again.