calibratelisted

# Calibrate — Config Editor (global + per-project) Read `${CLAUDE_PLUGIN_ROOT}/references/creed.md` and `${CLAUDE_PLUGIN_ROOT}/references/memory.md` first. Resolve memory first: `export SMITHY_MEM="$(bash ${CLAUDE_PLUGIN_ROOT}/scripts/paths.sh mem)"` — every smithy path below is relative to it, and it need NOT be inside the repo. If that dir does not exist, bootstrap per `${CLAUDE_PLUGIN_ROOT}/references/memory.md` § Location. Config has **three layers**, merged per key, lowest precedence first: | Layer | File | Scope | You may write it | |---|---|---|---| | defaults | `${CLAUDE_PLUGIN_ROOT}/defaults/config.json` | ships with the plugin | **NEVER** | | global | `$SMITHY_HOME/config.json` | every project on this machine | yes (`--global`) | | project | `$SMITHY_MEM/config.json` | this project only | yes (default) | You are the ONLY sanctioned writer of the global and project layers. Both stay SPARSE — they hold only what differs from the layer below. ## Process 1. **Show the current state.** Run: `bash ${CLAUDE_PLUGIN_ROOT}/scripts/routing.sh --dump` Present the table (role, model, effort, source). The SOURCE column names the layer that supplied each value — `defaults`, `global`, or `project`, with `(tier)` when a tier name expanded and `(translated)` when a cross-harness value was mapped by tier. The header also shows the resolved memory dir and how it was resolved. 2. **Parse one-shot arguments if given.** Accept `role=model/effort` pairs, e.g. `/