meme-engine

Structural review of a proof or derivation -- checks step justification, variable definitions, limiting cases, dimensional consistency, and canonical-source traceability

Quick symbolic/exact computation via the Sage MCP -- factor, simplify, symbolic eigenvalues, exact integrals. Use when float answers are not good enough

Classify a framework event, observable, or object as PROPAGATION (c-bounded, lives on g_M), SUBSTRATE DYNAMICS (not c-bounded, IS the substrate), MIXED, or CONTRADICTION -- walks a 6-step deterministic algorithm for emergent-metric / analog-gravity frameworks

Quick-access LaTeX typesetting expertise -- journal preambles, project macros, notation reference, document review, arxiv checklist

TikZ-based Penrose / Carter-Penrose diagram authoring -- preamble, pattern library, worked templates for Schwarzschild, Kerr, dS, CCC, dynamical compactification, and the M4 x SU(3)(tau) modulus-space transit. Use whenever a causal-structure diagram is needed.

Index the project's open conjectures/propositions -- scans sessions + open-question registers, tabulates with priority (EVOI), origin, related tests, and status

Create a new researcher agent with web-fetched research papers and agent definition

Execute a session plan in compute mode — hand each test case to its agent, wait, report. No teams, no inboxes.

Generate a workshop-schedule campaign from a just-closed session's working paper(s) -- find structural patterns (convergences, dissonances, uncovered corridors, elimination-without-explanation) and design the parallel solos (/rclab-review) + paired workshops (/rclab-workshop) + closeout. Workshop definitions and the workshop-vs-carry-forward discriminator come from `.claude/rules/Investigating-Workshops.md`; the schedule shape from `.claude/templates/workshop-schedule.md`. Small sources are read by one schedule planner; large multi-wave sources fan out to a per-chunk investigator swarm + one consolidator. Also supports an INVESTIGATION MODE (`--investigation n`) that mines an `investigation-{n}` effort's outputs into a synthesis + a forward seed for the next investigation and housekeeps `sessions/investigation/index.md`.

Plan the next compute session -- mechanical carry-forward gathering, wave partition, then a SWARM of per-wave planners that each design full-fidelity test cases for their slice of the carry-forward. Consolidate (single plan + single working paper) or fan-out (per-wave plan files + per-wave working papers). Also MAINTAINS and CONSUMES the project's forward registers (open-questions / priority / open-channel) from the just-mined session, so plan-time doubles as the session wrap-up. Supports an INVESTIGATION MODE (`--investigation [n] --from <seed>`) that plans an exploratory `investigation-{n}` effort in the parallel `sessions/investigation/` track from a free-form seed, as a MIXED-TYPE wave plan (compute / review / workshop gates). For workshop-schedule campaigns derived from the just-closed session, use `/rclab-investigate` instead.

Reflect on a just-closed wave, plan, session, or document -- what stood out, what was surprising, what patterns cross tests, what the next session should pursue. Summary plus introspection, not description.

Generate solo synthesis/review reports from source docs — 1+ independent agents each read the same sources and write their own report. No coordination between agents.

Execute a wave plan sequentially in the main agent session -- no subagent spawning. Task list is the primary driver: two tasks per gate (compute, update wp) in order.

Launch a coordinated multi-agent research team from a session file — TeamCreate, spawn with full assignments, inbox coordination, hands-off execution. For panel/interpretive and multi-round team-based collaborate formats where agents message each other by name. For independent parallel computation, use /rclab-coordinate. For 2-agent sequential document iteration, use /rclab-workshop.

2-agent iterative workshop on a shared document. Exactly 2 agents, N rounds (1-5), sequential — no team infrastructure. Each agent reads the running document, fills in their sections, completes before the next spawns.

Collapse and optimize agent memory files — deduplicate, compress, archive to docs

Query and maintain the project knowledge index

Scientific document builder — create papers, posters, slides, theses, reports from templates. Draft sections, peer review manuscripts, manage citations, compile LaTeX.

Read and process PDF files that exceed the Read tool's 10-page limit on Windows

Remove all agent-memory references to a session, mechanism, or identifier