claude-research

Use when you need to audit a research project against the init-project-research template.

Use when you need to validate a paper's bibliography — cross-references \cite{} keys against .bib files or embedded \bibitem entries, finds missing/unused/typo'd keys, and checks every key against the Paperpile library via the local resolver. Deep verification mode spawns parallel agents for DOI/metadata validation at scale. Fix mode rekeys drifted keys to canonical and stages missing entries for Paperpile.

Use when you need to bootstrap a full research project with directory scaffold and Overleaf symlink.

Use when you need to create a preprint / working-paper variant of a paper currently in conference or journal format. Forks the existing Overleaf project — adds a `preprint/` subfolder using the user's `your-template` Template, ports the body content from the source paper. The preprint is accessed locally via the existing `paper-{venue}/paper/preprint/` path (subfolder under the conference paper's symlink); no separate `paper-wp/` directory. Trigger on "set up a working paper", "create a preprint", "WP version", "arXiv-ready version", "ready to preprint". Never creates a new top-level Overleaf project — always nests inside the existing one. Never uses the conference's own style (.sty / .cls); always swaps to `your-template`.

Use when you need academic proofreading of a LaTeX paper (11 check categories).

Use when you need to assemble, anonymize, or audit a replication package.

Use when you need a mid-draft adversarial review of a paper — runs paper-critic + domain-reviewer + claim-verify + blindspot in parallel, then auto-synthesises into a prioritised revision plan. Distinct from /pre-submission-report (final-gate kitchen sink, 13 sub-agents) — this is the active-drafting feedback loop. Triggers: 'review my draft', 'adversarial review', 'cluster review', 'mid-draft critique', 'feedback before pre-submission'.

Use when mosh hangs / fails to connect to a headless Mac mini after a Tailscale update or restart, OR when a remote-access tool (RustDesk, VNC, AnyDesk) fails to reach its public relay from an MDM-managed client. Layers — Tailscale daemon health (dual-install conflict, headless GUI-app limitation, Homebrew launchd takeover), downstream mosh-server cleanup (stale UDP bindings to old Tailscale IP), macOS resolver stuck state, and Tailscale direct-IP as relay-bypass fallback. Symptoms include `could not get canonical name for <host>`, `failed to connect to local Tailscale service`, `Tailscale.CLIError error 1`, `Connection closed by UNKNOWN port 65535`, `Failed to connect to rs-ny.rustdesk.com:21116: Please try later`.

Use when you want to auto-promote project knowledge articles into vault wiki concepts. Saturday cron job — runs the corpus scanner, picks candidates above threshold, copies each source article into ~/vault/concepts/ with auto_generated frontmatter so it surfaces immediately on topic pages. Can also be run interactively for a dry-run preview.

Use when you need to scaffold a NEW educational companion book for a LaTeX paper. Reads the paper, drafts 8 substantive chapters into the vault at ~/vault/books/{slug}/, copies bib + figures, registers the book, and verifies atlas serves it. Source-of-truth is the paper PDF/tex; the book is a reading companion, never a re-statement of new claims. For syncing an existing book to a paper revision, use /audit-paper-book.

Use when you need to detect drift between an existing paper-book companion and a revised version of its source paper, then sync the mechanical pieces (new bib entries, new/changed figures) and report the substantive drift (renamed sections, changed numbers, new theorems, new contributions) for the user to triage. Counterpart to /init-paper-book. Read-only by default; --apply flag opts in to mechanical fixes.

Use when you need to extract citations from a PDF and generate a validated .bib file. Reads the PDF, identifies all referenced works, constructs BibTeX entries with metadata verification, then runs /bib-validate.

Use when you need to save session state to survive context compaction or handoff between sessions.

Use when you need to compile raw inputs (literature, meeting notes, session logs, code findings) into a per-project knowledge wiki. Supports --autonomous / -y for end-to-end runs without prompts (used by the Saturday wiki-grow cron).

Use when you need to gather PDFs from Paperpile into a project's articles/ folder.

Use when you want to be interrogated ONE question at a time — either to (a) defend your own research (viva, job talk, seminar Q&A, hostile-referee prep) or (b) study a class / subject you're learning (exam revision, active recall). An interactive adversarial/Socratic oral drill grounded in your actual material, escalating on weak answers, then a study sheet of what you fumbled with model answers. Triggers: 'grill me', 'grill me on X', 'viva prep', 'defend my paper', 'quiz me on this class', 'test me on <topic>', 'exam revision', 'help me study X'. Distinct from /devils-advocate (stress-tests arguments in prose) and referee2-reviewer / paper-critic (produce a WRITTEN critique) — here YOU answer aloud.

Use when you need to capture or integrate improvement ideas for Claude Code infrastructure.

Use when you need to check compiled knowledge for contradictions, uncited claims, missing connections, stale articles, and orphaned concepts.

Use when you need to MACHINE-CHECK a self-authored lemma/theorem by formalizing it in Lean 4 + mathlib and running `lake build` — the strongest verification (a clean build with no `sorry` IS a proof). R3 of the verification spectrum. Runs against the pre-seeded mathlib project at ~/lean-verify/mathlib_verify on the Mac Mini. Triggers: /lean-check, 'formalize this lemma in Lean', 'machine-check this theorem', 'prove this in Lean'. NOT for stress-testing a distributional claim (use /numerical-check), an algebra step (use /symbolic-check), or claims too rich to faithfully state (escalate to domain-reviewer). The hard part is a FAITHFUL statement — a Lean lemma that doesn't match the paper is false confidence.

Use when you need academic literature discovery, synthesis, or bibliography management. Supports standalone searches and end-to-end project pipelines with vault sync and auto-commit.

Manage old recordings — find large files, archive old meetings, delete processed originals. Use when the user says "clean up recordings", "how much space are meetings using", "delete old recordings", "archive meetings", "manage meeting storage", or asks about disk space from minutes.

Post-meeting debrief — analyzes what happened, compares outcomes to your prep intentions, tracks decision evolution. Use when the user says "debrief", "what just happened in that meeting", "what did we decide", "debrief that call", "post-meeting", "what changed", or right after stopping a recording.

List recent meetings and voice memos. Use when the user asks "what meetings did I have", "show my recent recordings", "any meetings today", "list my voice memos", or wants an overview of their meeting history. Also use when they need to find a specific meeting by browsing rather than searching.

Interactive meeting preparation — builds a relationship brief and talking points before a call. Use when the user says "prep me for my call with", "I'm meeting with X", "prepare me for", "what should I bring up with", "meeting prep", "get ready for my call", or wants to review history with someone before a meeting.

Generate a daily digest of today's meetings and voice memos — key decisions, action items, and themes across all recordings. Use when the user asks "recap my day", "what happened in my meetings today", "daily summary", "what did I discuss today", "any action items from today", or wants a consolidated view of the day's conversations.

Search past meeting transcripts and voice memos for specific topics, people, decisions, or ideas. Use this whenever the user asks "what did we discuss about X", "find that meeting where we talked about Y", "what did Alex say", "did we decide on", "what was that idea about", or any question that could be answered by searching their meeting history. Also use for "do I have any notes about" or "check my meetings for".

Verify that the meeting-transcribe pipeline is properly set up — whisper/ffmpeg/uv binaries, models, package venv, watcher, and folders. Use when the user says "is transcription working", "check my meeting setup", "verify minutes", "why isn't transcription working", "meeting health check".

Weekly meeting synthesis — themes, decision arcs, stale commitments, and what deserves your attention next week. Use when the user says "weekly review", "what happened this week", "weekly summary", "recap my week", "any outstanding items", "week in review", or at the end of a work week.

Use when the user needs to write a preregistration for an empirical study before collecting data — OSF-Standard 'Pre-Data Collection Registration' by default, or the shorter AsPredicted 9-question form via `--aspredicted`. Triggers on phrases like 'preregister this study', 'OSF prereg', 'AsPredicted', 'register hypotheses', 'prereg the experiment', 'lock in the analysis plan before running', or whenever the user is about to launch a Prolific / lab / survey / computational-audit experiment and wants a time-stamped record of the design + analysis. Produces a single, upload-ready markdown file with every standard section filled in (hypotheses, design, sampling, variables, analysis, exclusions, stopping rule). Composes /hypothesis-generation, /experiment-design, /causal-design, /synthetic-data (power), and /ethics-review by reference.

Use when you need to rename a research project across all systems (local directory, atlas, vault, Overleaf, git).

Use when you need content-aware file reorganisation or deduplication driven by a LOCAL model (Ollama on the Mac Mini) that actually reads each file's contents, not just its name — proposes a folder structure or finds exact/near duplicates, always dry-runs first, and executes only on approval with a full undo. Triggers: 'reorganise this folder', 'tidy up Downloads with the local model', 'reorg', 'find duplicate files', 'dedup this folder', 'sort these files by reading them'.

Use when you need to retarget a paper to a different journal (rename, swap bib, update citations).

Use when you need to run the end-of-session checklist (uncommitted changes, focus update, project sync). Auto-detects research projects and adds atlas consistency checks.

Use when you need a dashboard of skill invocation counts, success rates, failure patterns, and health status. Reads JSONL logs and qualitative observations. Reporting only — never auto-modifies skills.

Use when you need to sync global permissions into the current project.

Use when you need help with daily planning, weekly reviews, meeting actions, or vault task queries.

Use when you need to audit the vault wiki (~/vault/concepts/) for fragmentation, missing tags, write-only concepts, and draft/anatomy conformance. Read-only — produces a markdown report at /tmp/wiki-curate-report.md. Companion to /wiki-grow (which writes) and /wiki-merge (which fixes overlap clusters).

Use when /wiki-curate has identified an overlap cluster and you want to collapse it. Takes a canonical winner slug + one or more fold-in slugs; merges bodies, rewrites all [[wikilink]] references in the corpus, deletes losers, denylists them. Dry-run by default; --apply writes.

Use when you need to wire a shared Python package as an editable dependency across projects.

Use when you need to check a LaTeX submission against a PDF assessment brief.