ki-tokenomicslisted

Codify and audit portable agent-context tokenomics: repository-selected standing-surface attribution, budget guide-rails, portable model-purpose taxonomy, and the `[skills.ki-tokenomics]` configuration table. Use when a repository needs a runtime-neutral context-cost policy, model-purpose choice, or token budget. Triggers: "set a context budget", "audit our tokenomics policy", "which model type should this work use", "configure tokenomics". Runtime evidence belongs to `ki-tokenomics-claude` or `ki-tokenomics-codex`; MCP server design belongs to `ki-repo-mcp`; skill-description quality belongs to `ki-skills`.
knowledgeislands/ki-agentic-harness · ★ 0 · AI & Automation · score 76

Install: claude install-skill knowledgeislands/ki-agentic-harness

# Knowledge Islands tokenomics `ki-tokenomics` owns the portable policy for the context a runtime carries on every turn. The policy is deliberately separate from inspecting a vendor's files: `ki-tokenomics-claude` and `ki-tokenomics-codex` compose this skill and supply their documented, bounded runtime evidence. The standing surface is the selected repository's instructions, memory, installed-skill descriptions, MCP tool definitions, and settings or output configuration. An audit attributes each measured cost to its repository or runtime-user layer and routes an artifact fix to its owner: `ki-repo-mcp` for an MCP server, `ki-skills` for a skill description, and the relevant runtime adapter for runtime configuration. Budgets are guide-rails, not gates: every measured budget overage is a **WARN**, never a FAIL. FAIL is reserved for a malformed selected-repository `[skills.ki-tokenomics]` table or an explicitly required runtime integration that is missing. Estimates are labelled `~`; they are not billing or provider token counts. ## Portable configuration The selected repository may declare only this skill's table. It is read validate-down: unknown keys WARN, recognised keys must have the documented shape, and no other skill table is inspected. ```toml [skills.ki-tokenomics] headroom = "recommended" # "required" | "recommended" | "off" context_window_tokens = 200000 preferred_model_type = "standard" # frontier | reasoning | standard | fast [skills.ki-tokenomics.model_tier