ARIA-NBV

Use for ARIA-NBV Typst proposal/thesis authoring, shared notation, scientific prose, citations, scientific/geometric figures, tables, Mermaid inclusion, and compile/render QA.

Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community detection, and query/path/explain tools.

Use to localize unknown ARIA-NBV files, symbols, docs, or source families through deterministic local discovery before handoff.

Use when reviewing ARIA-NBV working-tree or PR diffs, following up external/model review findings, or supplying ARIA-NBV domain context to the oh-my-codex code-review harness with severity-ranked file and line findings.

Use to diagnose ARIA-NBV bugs, regressions, failing metrics, Streamlit issues, docs builds, KG failures, or suspicious outputs.

Use when ARIA-NBV work touches pose, camera, coordinate-frame, CW90, PyTorch3D projection, depth backprojection, candidate frusta, or geometry diagnostics contracts.

Stress-test vague, high-impact, research-facing, advisor-facing, or cross-surface ARIA-NBV decisions before implementation, including theory-rich or elaborate Plan Mode option analysis when requested.

Use for ARIA-NBV behavior-preserving pruning of redundancy, dead code, stale compatibility, unused config, or excess LOC.

Write and refactor concise, contract-focused Python docstrings for modules, public classes, functions, methods, protocols, config models, DTOs, wrappers, and streaming or session APIs. Use when Python docstrings are missing, sparse, misleading, or need better cross-references, examples, units, shapes, lifecycle notes, or boundary semantics.

Use when operating or documenting ARIA-NBV ASE downloads, ATEK shards, meshes, immutable VIN offline stores, dataset-version updates, split manifests, storage estimates, and data smoke checks.

Use when working with LRZ AI Systems remote compute for ARIA-NBV: SSH/login.ai.lrz.de, DSS storage, Slurm GPU/CPU allocations, Enroot/Pyxis containers, dataset/cache/training batch jobs, or debugging remote job failures.

Use when ARIA-NBV work changes Zarr-Python API usage, chunking, codecs, stores, sharding, concurrency, or v2/v3 migration behavior in offline or rollout storage.

Use before non-trivial ARIA-NBV work to choose a lane, state assumptions, inspect owners, keep diffs traceable, and verify.

Use when triaging or maintaining ARIA-NBV internal agent-memory and backlog TOML surfaces with `make agents-db`.

Use for KG-backed ARIA-NBV retrieval, task routing, claim checks, current-truth checks, backlog lookup, and consolidation proposals.

Use for creating, editing, validating, or rendering ARIA-NBV Mermaid `.mmd` thesis diagrams and diagram templates.

Use when ARIA-NBV work touches ASE counterfactual rollouts, non-myopic planning evaluation, invalid-action handling, stochastic branches, finite-candidate candidate-query Transformer Q_H, or the roadmap value/RL gate. Gymnasium/SB3 is post-M6 bridge work only.

Use when ARIA-NBV work changes public docs, Typst/Quarto narrative, bibliography, navigation, or the public/internal docs boundary.

Use when ARIA-NBV work touches target/entity selection, GT OBB crops, target-specific RRI labels, target-conditioned VIN fields, or entity-aware diagnostics.

Use when creating, reviewing, or fixing ARIA-NBV Rerun integrations for immutable VIN offline-store inspection, NBV candidate/frustum visualization, RRI/validity diagnostics, depth/RGB/keyframe layers, OBB/mesh/trajectory logging, `.rrd` smoke artifacts, or Rerun frame-coordinate issues involving PoseTW, CameraTW, PyTorch3D cameras, and display-only CW90 handling.

Use when changing or operating ARIA-NBV litkg-rs ingestion, KG config, backend/export contracts, source adapters, or generated KG artifacts.

Evaluate and plan incremental Mojo adoption for Aria-NBV when users ask about Mojo, Modular, kernel ports, GPU acceleration, FFI boundaries, or whether a specific `aria_nbv` hot path should move out of Python or PyTorch. Use for repo-specific decisions about what to port, how to preserve current interfaces, and when to keep code in Python.