@alecsibilia/luca 13.0.0 → 13.0.1

package/dist/claude/.claude/luca-statusline.ts

@@ -74,8 +74,21 @@ async function readContextTokens(transcriptPath) {
   } catch {}
   return null;
 }
+var ONE_MILLION_MODEL_IDS = [
+  "fable",
+  "mythos",
+  "opus-4-6",
+  "opus-4-7",
+  "opus-4-8",
+  "sonnet-4-6",
+  "sonnet-5"
+];
 function contextLimit(modelId) {
-  return modelId !== undefined && modelId.includes("[1m]") ? 1e6 : 200000;
+  if (modelId === undefined)
+    return 200000;
+  if (modelId.includes("[1m]"))
+    return 1e6;
+  return ONE_MILLION_MODEL_IDS.some((id) => modelId.includes(id)) ? 1e6 : 200000;
 }
 function formatTokens(n) {
   if (n >= 1e6) {

package/dist/claude/skills/lu/SKILL.md

@@ -64,7 +64,15 @@ Triage runs once, at the start of a run. It is inline here — there is no separ
 
 1. **Classify complexity.** Read the request. Pick one of `TRIVIAL | SIMPLE | MODERATE | COMPLEX | CRITICAL` based on file count, scope, and risk. There is no CLI command to persist complexity — record it in your reasoning and pass it to every subagent you spawn (pass it to any subagent whose behavior varies by complexity). If `--complexity=<level>` or `--force-complex` was passed, use that directly.
 
-   **Override telemetry.** Once the final complexity level is chosen, compare it against the deterministic heuristic baseline from `luca classify --task "<request>" --json` (the `--task` flag is REQUIRED — read its `.complexity` field). On a MISMATCH between the heuristic level and your final level, emit one override record:
+   **Heuristic baseline (floor, not ceiling).** Compute the deterministic baseline with `luca classify`, passing the scope signals you already extracted while reading the request — NOT just `--task`. A `--task`-only call starves the heuristic (it scores `estimatedFileCount=0`, no domains, no concerns) and collapses to description-keyword matching, which systematically under-scores work. Supply every signal you can estimate:
+   ```
+   luca classify --task "<request>" --files <estimated-file-count> --domains "<comma-list>" --concerns "<comma-list>" [--breaking] --json
+   ```
+   (`--task` is REQUIRED; `--files`/`--domains`/`--concerns`/`--breaking` are optional — pass each one you can estimate. Read the `.complexity` field.)
+
+   This baseline is a **FLOOR, never a ceiling.** Breadth signals (files, domains) cannot measure design/iteration depth — a deep single-file design, tuning, or balancing task is genuinely hard yet touches one file. Your own judgment is authoritative: take the HIGHER of the heuristic level and your read, and never demote below your judgment to match the heuristic. **In particular, never trust a `TRIVIAL` heuristic result on its face — re-judge it before accepting**, because a breadth-only score under-rates deep single-file work.
+
+   **Override telemetry.** On a MISMATCH between the heuristic level and your final level, emit one override record:
    ```
    luca telemetry emit --kind classifier.override --run-id <runId> --meta '{"classifier":"complexity","from":"<heuristic-level>","to":"<final-level>","source":"<taxonomy>"}'
    ```

package/dist/index.mjs

@@ -2,7 +2,7 @@ import { defineCommand, runMain as runMain$1 } from 'citty';
 import 'pathe';
 
 async function resolveVersion() {
-  return "13.0.0";
+  return "13.0.1";
 }
 const LUCA_VERSION = await resolveVersion();
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@alecsibilia/luca",
   "description": "Luca — spec-driven agentic development workflow for Claude Code (umbrella; bundles luca-cli, luca-core, luca-tools)",
-  "version": "13.0.0",
+  "version": "13.0.1",
   "license": "Apache-2.0",
   "author": "Alec Sibilia <sibilia.alec@gmail.com>",
   "homepage": "https://github.com/asibilia/luca-framework#readme",