kb-llm-strengths

name	kb-llm-strengths
description	The model-tiering knowledge base — which model to use for planning, implementation, and verification. Read a kb/role-*.md file when you are about to dispatch or pick a model for a task.

LLM strengths — the model-tiering knowledge base

This package is a knowledge base (a kb/ subfolder, declared by the [kb] marker in lanius.toml). It holds the model-tiering rules the harness uses when it dispatches work: which model plans, which implements, which verifies, and the one rule that never flexes.

It is deliberately mutable and preference-dominated: benchmark claims are the floor, Tim's lived experience overrides them. When they conflict, human preference wins and the conflict is flagged, not silently resolved.

How this KB is laid out

• One file per role — start here when you are dispatching:
  • kb/role-planner.md — who plans (and who must never)
  • kb/role-implementer.md — who implements
  • kb/role-verifier.md — who verifies
• One file per model — the per-model notes the role files link into:
  • kb/claude.md, kb/fable.md, kb/opus.md, kb/gpt-5.5.md, kb/glm-5.2.md

Using it

• Read a kb/role-*.md before you pick a model for a task, or grep the tree: grep -ri "who verifies" kb/.
• List every enabled KB with lanius kb list.
• Write an update with lanius kb write kb-llm-strengths kb/<file>.md (it writes the file and commits it — provenance is the git log). Update the paired copy too: the canonical model-tiering text is mirrored in .claude/skills/handoff-workflow/SKILL.md, which configures Claude Code (and cannot read this KB). Keep the two from forking.