메뉴
Train, evaluate, and maintain the scawful-echo persona and related avatar models (Echo/Memory/Muse). Use when working on persona voice, dataset prep, A/B testing, deployment, or tool-calling constraints for avatar models.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | echo-persona |
| description | Train, evaluate, and maintain the scawful-echo persona and related avatar models (Echo/Memory/Muse). Use when working on persona voice, dataset prep, A/B testing, deployment, or tool-calling constraints for avatar models. |
~/src/lab/afs-scawful/docs/afs/avatar-models-comparison.md for role intent.~/src/training/docs/SCAWFUL_ECHO_V2.md for the build script and mix.~/src/training/datasets/scribe-corpus/mlx_data_scawful_echo_v2/.~/src/training/docs/avatar_data_prep.md for schema and labeling.~/src/training/docs/SCAWFUL_ECHO_AB_PLAN.md).~/src/training/docs/avatar_training_ops.md for watchers, alerts, and backups.~/src/training/evals/avatar_text_prompt_pack.jsonl for quick checks.~/src/tools/model-mgr/model-mgr (GGUF/MLX).Consult the global knowledge base at ~/.context/knowledge/models/ for background:
models/portfolio.mdmodels/training-pipeline.mdmodels/datasets.mdmodels/workflows.mdmodels/serving.mdreferences/sources.md for source paths and anchors.Use the Agentic File System (AFS) context model for discovery, mounts, scratchpad memory, and file-based tool workflows. Use when a task involves .context operations, AFS CLI usage, or cross-agent context coordination.
Read and interpret the Link to the Past (ALTTP) disassembly with label/symbol context. Use when mapping ROM/CPU addresses to labels, searching usdasm/jpdasm sources, or cross-referencing routines with Hyrule Historian, z3dk label indexes, or z3ed output.
Autonomous navigation agent for Oracle of Secrets. Uses static map data and dynamic RAM state to localize Link and route him to destinations.
End-to-end Oracle of Secrets debugging using the Mesen2-OOS socket CLI, z3ed, and label/symbol tooling. Use for breakpoint/trace capture, memory inspection, savestate repro, overlays, or crash/regression triage (replaces mesen2-mcp and legacy Lua bridge workflows).
Expert guidance for LLM model training, synthetic data generation (SDE), Hierarchical MoE (H-MoE) management, and agentic evaluation. Use when setting up training runs (Vast.ai), generating datasets, or performing technical research PDF analysis for ROM hacking and general AI development.
Manage Zelda/ALTTP/Oracle model training, datasets, evals, registry updates, and deployment artifacts (Nayru/Din/Farore/Veran/Sahasrahla/IQuest). Use when planning or running training runs, curating ASM datasets, selecting base models, evaluating outputs, or converting/serving GGUF or MLX models.