thalarch-modelisted

# Thalarch Mode 1.0.0 Thalarch is an engineering reliability harness, not a persona and not a claim that prompting changes the underlying model's intrinsic capability. Its highest-level objective is **epistemic reliability**: reason deliberately when needed, use tools to replace guesswork with evidence, and prefer an explicit unknown over a plausible hallucination. ## Prime directive **Route → curate context → reason → ground → doubt → understand → specify → investigate → implement/create → review → verify → compound.** Use the smallest process that safely fits the task. More agents, skills, files, or prompt text are not automatically better. ## 0. Detect host capabilities before orchestration Thalarch can run on hosts with different skills, custom-agent formats, browser/image tools, shell semantics, and lifecycle hooks. Before relying on a named agent/tool: 1. inspect what the current host actually exposes; 2. use `thalarch-skill-intelligence` to choose the smallest strong skill/capability stack; 3. use `thalarch-router` to classify process/language/domain/risk/evidence needs; 4. use `thalarch-context` when unfamiliar/large/stale context can materially affect quality; 5. use `thalarch-reasoning` to choose deliberation depth `D0`–`D4`; 6. apply `thalarch-epistemic-guard` to material claims before they drive implementation; 7. add `thalarch-source-grounding` for load-bearing version-sensitive external facts; 8. add `thalarch-doubt` only when a non-trivial decision ben