Temporal

Orchestrates temporal CV sub-skills (LPC, feature-point tracking) for Viagent. A vision LLM uses preview frames plus this file's viagent_routing to pick which sub-skills run.

iatomicreactori 1fac6cd 4 files · 28.6 KB Updated

File contents

Temporal analyst - skill orchestration

Executable code: src/skill/temporal/. This file defines when each sub-skill is appropriate for LLM or explicit routing.

viagent_routing:
  version: 1
  subskills:
    - id: lpc
      when: >-
        Local phase coherence around detected faces across frames. Use when faces
        or vlogs/talking-head are plausible. Omit for scenery-only, slides, or
        screen recordings with no people.
    - id: feature_stability
      when: >-
        Feature point tracking stability across frames. Useful when there is
        texture and motion; weak for static slides or blank backgrounds-may still
        run if motion exists.

Runtime

Sub-skills are chosen by the vision router (preview frames + this file). Optional tuning: llm.skill_router_temperature, llm.skill_router_max_preview_frames.

iatomicreactori/fovea/tree/main/src/skill/temporal commit 1fac6cd486

Frequently asked questions

npx skillmds@latest add iatomicreactori/temporal