vivariumlisted

# vivarium — durable comparative-genomics orchestrator Coordinate multi-stage comparative-genomics analyses. For new end-to-end work, use the 2.0 event-sourced engine so the append-only ledger is the source of state authority. Run eligible stages locally; pause at scaffold or cluster stages and resume from the same ledger after outputs return. This skill coordinates; analysis implementations live in the sub-skills. Read the relevant sub-skill before executing its stage. | Sub-skill | Does | |---|---| | `vivarium-prep` | assembly QC (stats, CheckM2), annotation (Prokka → eggNOG/dbCAN) | | `vivarium-compare` | ANI/AAI, orthology (OrthoFinder), synteny (MUMmer) | | `vivarium-phylo` | alignment → trim → tree (IQ-TREE), selection (PAML dN/dS) | | `vivarium-search` | BLAST/DIAMOND/HMMER sequence search | | `vivarium-report` | standardized manuscript figures and tables (Python/R) | ## Select the execution mode - **Route by scale, not by habit.** The durable kernel earns its token and artifact overhead only past a threshold. Drive it (`plan`/`run`, `full`) when the project is long-running, accumulates more state than fits one context window, must survive a crash, or needs an auditable commit chain or cluster routing. For a one-shot analysis that fits comfortably in context (a single ANI, one tree, one figure, or a short chain), run the stage(s) directly via the sub-skills — cheaper, with no correctness penalty. Benchmark basis: on a context-sized task the kernel added tokens (≈