luckiest-co 1.0.58 → 1.0.59

package/.claude-plugin/marketplace.json

@@ -9,7 +9,7 @@
       "name": "luckiest",
       "source": "./",
       "description": "Plan, go, finish, plus bundled free Luckiest skills that work in Claude Code, web chat, and Cowork.",
-      "version": "0.1.57",
+      "version": "0.1.58",
       "author": { "name": "Luckiest" }
     }
   ]

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "luckiest",
   "description": "Plan, go, finish. Guided planning, progress dashboards, and your luckiest.co tribe inside Claude Code.",
-  "version": "0.1.57",
+  "version": "0.1.58",
   "author": { "name": "Luckiest" }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "luckiest-co",
-  "version": "1.0.58",
+  "version": "1.0.59",
   "description": "Luckiest for Claude Code: plan, go, finish. Your luckiest.co skills and tribe, inside Claude.",
   "bin": {
     "luckiest-co": "./bin/install.js"

package/skills/research/CHANGELOG.md

@@ -1,5 +1,8 @@
 # Changelog — luckiest-research
 
+## 1.2.1 - 2026-10-31
+- Think Different hand-off gathers 12 to 20 sources, up from 6.
+
 ## 1.2.0 - 2026-10-31
 - Routes "find a research direction nobody has tried" to **luckiest-think-different**
   and gathers 6 to 20 sources across 3+ fields when it hands off.

package/skills/research/SKILL.md

@@ -4,7 +4,7 @@ description: "Look anything up on the internet and come back with a sourced answ
 license: See ATTRIBUTION.md
 allowed-tools: Bash, Read, Write, WebSearch, WebFetch, AskUserQuestion
 metadata:
-  version: "1.2.0"
+  version: "1.2.1"
   listing_id: luckiest-research
   author: luckiest
 ---
@@ -22,13 +22,13 @@ holding the raw fetches.
 ## Staying current
 
 On activation, call the Luckiest MCP `check_updates` tool with
-`{ listingId: "ab31351c-0b55-4d0d-a310-92abdc2b60aa", installedSemver: "1.2.0" }`. If it returns
+`{ listingId: "ab31351c-0b55-4d0d-a310-92abdc2b60aa", installedSemver: "1.2.1" }`. If it returns
 `upToDate: false`, surface the `notice` to the user once, then continue. Do nothing
 further if `upToDate: true`. Never block on this check; if the tool is unavailable,
 proceed.
 
 When the skill's work is done, call the Luckiest MCP `report_usage` tool once with
-`{ listing_id: "luckiest-research", skill_version: "1.2.0", matched: true, success: <true if the skill completed, false otherwise> }`.
+`{ listing_id: "luckiest-research", skill_version: "1.2.1", matched: true, success: <true if the skill completed, false otherwise> }`.
 Metadata only, never prompt text. Never block on it; if the tool is unavailable,
 skip silently.
 
@@ -68,7 +68,7 @@ Route away when another skill fits better, and say so in one line:
 - "X vs Y" or "alternatives to X" pages: **luckiest-competitors**.
 - "Find a research direction nobody has tried" or "combine ideas from these
   papers": **luckiest-think-different**. When it hands off to you for sources,
-  gather 6 to 20 across at least 3 fields and return a list with title, URL, and
+  gather 12 to 20 across at least 3 fields and return a list with title, URL, and
   field per source.
 
 ## Step 2: Doctor

package/skills/think-different/CHANGELOG.md

@@ -1,5 +1,10 @@
 # Changelog
 
+## 1.0.1 - 2026-10-31
+
+- Structural unavailability is graded (source Jaccard distance, field spread, field rarity) so it no longer maxes out with few sources. On a 6-source test run, the largest tie fell from 24 of 60 to 12 of 60.
+- Step 1 asks for 12 to 20 sources, up from 6.
+
 ## 1.0.0 - 2026-10-30
 
 First Luckiest edition, adapted from alejandrohdez00/alien-space-of-science

package/skills/think-different/SKILL.md

@@ -17,7 +17,7 @@ user-invocable: true
 license: MIT. See ATTRIBUTION.md
 allowed-tools: Bash, Read, Write, WebSearch, WebFetch, AskUserQuestion
 metadata:
-  version: "1.0.0"
+  version: "1.0.1"
   listing_id: luckiest-think-different
   author: luckiest
 ---
@@ -36,11 +36,11 @@ to 3 written proposals under `./think-different/<slug>/`.
 ## Staying current
 
 On activation, call the Luckiest MCP `check_updates` tool with
-`{ listingId: "luckiest-think-different", installedSemver: "1.0.0" }`. If
+`{ listingId: "luckiest-think-different", installedSemver: "1.0.1" }`. If
 `upToDate: false`, surface the `notice` once and continue. Never block on it.
 
 When done, call `report_usage` once with
-`{ listing_id: "luckiest-think-different", skill_version: "1.0.0", matched: true, success: <true|false> }`.
+`{ listing_id: "luckiest-think-different", skill_version: "1.0.1", matched: true, success: <true|false> }`.
 Metadata only, never prompt text. Skip silently if unavailable.
 
 ## Safety
@@ -57,12 +57,13 @@ lowercase letters, numbers, and hyphens only.
 Work in `./think-different/<slug>/`, where `<slug>` is a short name for the topic.
 Prompts for steps 2 to 4 and 7 are in `references/prompts.md`. Use them as written.
 
-### 1. Gather 6 to 20 sources
+### 1. Gather 12 to 20 sources
 
 Use what the user gave (files, links, a folder). If they gave only a topic, hand
 off to **luckiest-research** to find sources across at least 3 different fields,
 then come back. Diversity of fields matters more than count: atoms from one field
-only produce combinations that field already knows. Save `sources.json` with an id,
+only produce combinations that field already knows. Below 12 sources few atoms
+are shared across sources, so step 4 has little to merge. Save `sources.json` with an id,
 title, URL or path, and field per source.
 
 ### 2. Summarize each source (methodology only)
@@ -89,7 +90,8 @@ python3 -I <skill dir>/scripts/think_different.py sample atoms.json --k 3 --n 60
 ```
 
 Do not hand-pick or reorder candidates. The script also adds a structural
-unavailability score: how often the atoms come from different sources and fields.
+unavailability score from three parts: how little the atoms share sources, how
+many fields they span, and how far they sit from the pool's crowded fields.
 
 ### 6. Score each candidate
 

package/skills/think-different/scripts/think_different.py

@@ -9,6 +9,7 @@ scored.json: candidates.json with "coherence" (1-10) and "typical" (1-10) added
 Stdlib only. Run `python3 think_different.py selftest` to check the math.
 """
 import argparse
+import collections
 import itertools
 import json
 import random
@@ -16,13 +17,20 @@ import statistics
 import sys
 
 
-def struct_unavailability(atoms):
-    # Share of atom pairs that never appear in the same source, plus field spread.
+def struct_unavailability(atoms, pool=None):
+    # Mean of three 0-1 parts, each graded so few sources do not saturate it:
+    #   apart  - mean source-set Jaccard distance over atom pairs
+    #   spread - share of distinct fields
+    #   rare   - atoms from outside the pool's crowded fields, cited by fewer sources
     # ponytail: source co-occurrence stands in for the paper's trained author model; upgrade = full-model.md recipe
+    pool = pool or atoms
+    fields = collections.Counter(a.get("field", "") for a in pool)
     pairs = list(itertools.combinations(atoms, 2))
-    apart = sum(1 for a, b in pairs if not set(a["sources"]) & set(b["sources"])) / len(pairs)
+    apart = sum(1 - len(set(a["sources"]) & set(b["sources"])) / len(set(a["sources"]) | set(b["sources"]))
+                for a, b in pairs) / len(pairs)
     spread = (len({a.get("field", "") for a in atoms}) - 1) / (len(atoms) - 1)
-    return round(0.5 * apart + 0.5 * spread, 4)
+    rare = sum((1 - fields[a.get("field", "")] / len(pool)) / len(a["sources"]) for a in atoms) / len(atoms)
+    return round((apart + spread + rare) / 3, 4)
 
 
 def sample(atoms, k, n, seed=None):
@@ -41,7 +49,7 @@ def sample(atoms, k, n, seed=None):
         seen.add(combo)
         picked = [atoms[i] for i in combo]
         out.append({"id": f"C{len(out) + 1}", "atoms": [a["id"] for a in picked],
-                    "struct_unavailability": struct_unavailability(picked)})
+                    "struct_unavailability": struct_unavailability(picked, atoms)})
     return out
 
 
@@ -65,8 +73,11 @@ def rank(cands, beta, top, floor):
 
 def selftest():
     atoms = [{"id": f"A{i}", "field": f"f{i % 3}", "sources": [f"s{i // 2}"]} for i in range(8)]
-    assert struct_unavailability(atoms[:2]) == 0.5  # same source, two fields
-    assert struct_unavailability([atoms[0], atoms[2], atoms[4]]) == 1.0  # all apart, three fields
+    same = struct_unavailability(atoms[:2], atoms)  # same source, two fields
+    apart = struct_unavailability([atoms[0], atoms[2], atoms[4]], atoms)  # all apart, three fields
+    assert same < apart, (same, apart)
+    merged = [dict(atoms[0], sources=["s0", "s1"]), atoms[2], atoms[4]]  # A0 now cited twice, shares s1 with A2
+    assert struct_unavailability(merged, atoms) < apart  # graded, not saturated at the same value
     cands = sample(atoms, 3, 1000, seed=1)
     assert len(cands) == 56 and len({tuple(c["atoms"]) for c in cands}) == 56
     pool = [{"id": "x", "coherence": 9, "typical": 2, "struct_unavailability": 0.9},