スキル一覧に戻る
Folly-Partners

superthings

by Folly-Partners

Syncs my Claude code instances across computers.

0🍴 0📅 2026年1月25日
GitHubで見るManusで実行

SKILL.md


name: superthings description: Use when triaging Things inbox, creating tasks, or managing todos. Learns from corrections to improve title suggestions and project assignments over time.

SuperThings - Intelligent Task Management

SuperThings learns from your corrections to get better at suggesting titles and projects for Things tasks.

Learning System (Agent-Native)

The learning system is fully agent-native via MCP tools. Agents can read, write, and query learned patterns programmatically.

MCP Tools for Learning

ToolPurpose
things_list_patternsList all learned patterns (title_transforms, project_hints, exact_overrides)
things_suggest_for_taskGet suggestion + confidence for a task title
things_log_correctionLog a single correction to teach the system
things_update_patternAdjust pattern confidence manually
things_remove_patternRemove a bad or outdated pattern
things_learn_batchProcess multiple triage decisions at once

Example Agent Workflow

An agent can now triage inbox programmatically:

  1. things_get_inbox - Get inbox items
  2. For each item: things_suggest_for_task - Get suggestions with confidence
  3. If confidence < 3: Use AskUserQuestion with context
  4. If confidence >= 3: Auto-apply with [learned: Nx] indicator
  5. things_learn_batch - Update patterns from decisions
  6. things_update_todo - Apply changes to Things

Pattern Storage

Patterns stored in ~/claudesync/servers/super-things/data/patterns.json:

{
  "title_transforms": [
    {
      "match": "^Fix ",
      "transform": "Delegate to Brianna: {original}",
      "confidence": 12,
      "examples": ["Fix fireplace", "Fix garage door"],
      "last_used": "2026-01-15T10:00:00Z"
    }
  ],
  "project_hints": {
    "build": {"Deep Work": 15, "Computer": 2},
    "research": {"Computer": 23, "Deep Work": 1},
    "call": {"Call": 18}
  },
  "exact_overrides": {
    "Call mom": {"title": "Call: Mom", "project": "Call", "confidence": 8}
  },
  "stats": {
    "sessions_completed": 47,
    "items_processed": 892,
    "patterns_learned": 23,
    "accuracy_trend": [0.60, 0.72, 0.85]
  }
}

History Logging

All corrections logged to ~/claudesync/servers/super-things/data/history.jsonl:

{"ts": "2026-01-20T10:00:00Z", "original": "Fix fireplace", "suggested_title": "Delegate to Brianna: Fix fireplace", "final_title": "Delegate to Brianna: Fix fireplace", "suggested_project": "Computer", "final_project": "Computer", "title_accepted": true, "project_accepted": true}

Confidence Thresholds

ConfidenceBehavior
< 3Use AskUserQuestion with context (pattern, examples)
>= 3Auto-apply with [learned: Nx] indicator
>= 10Apply silently (trusted pattern)

AskUserQuestion for Low Confidence

When confidence is below 3, present options with full context:

question: "Title suggestion for 'Fix fireplace'?"
header: "2x"
options:
  - label: "Delegate to Brianna: Fix fireplace"
    description: "Pattern: ^Fix → Delegate | 2x confidence | Examples: Fix garage, Fix dishwasher"
  - label: "Fix fireplace"
    description: "Keep original unchanged"

Using the Learning Tools

Reading Patterns

things_list_patterns
→ Returns all title_transforms, project_hints, exact_overrides with counts

Getting Suggestions

things_suggest_for_task({ title: "Fix dishwasher" })
→ {
    original: "Fix dishwasher",
    suggested_title: "Delegate to Brianna: Fix dishwasher",
    suggested_project: "Computer",
    confidence: 12,
    pattern_source: "title_transform",
    pattern_rule: "^Fix ",
    examples: ["Fix fireplace", "Fix garage door"]
  }

Logging Corrections

Single correction:

things_log_correction({
  original_title: "Fix fireplace",
  suggested_title: "Delegate to Brianna: Fix fireplace",
  final_title: "Delegate to Brianna: Fix fireplace",
  title_accepted: true,
  project_accepted: true
})

Batch of corrections:

things_learn_batch({
  decisions: [
    { original_title: "Fix X", final_title: "Delegate: Fix X", title_accepted: true },
    { original_title: "Research Y", final_project: "Computer", project_accepted: true }
  ]
})

Managing Patterns

Update confidence:

things_update_pattern({
  pattern_type: "title_transform",
  key: "^Fix ",
  confidence_delta: 1  // or set absolute with confidence: 10
})

Remove bad pattern:

things_remove_pattern({
  pattern_type: "title_transform",
  key: "^Fix "
})

Projects Reference

ProjectID
ComputerLDhUsibk3dp2ZPioQySSiu
Deep WorkWamuBi2sFwbUwpXz9NZetP
CallEr67bc9YAur6ZeKTCBLC4c
HomeNsSR9HR3pd2bVi2z4QHFfM
Out and About5LwYiPJAkWGSCMHML8xaXb
Kids/Activities7n14Jusf7nCrLADfAANabW
SomedayEZ5uJWRtcrvJ4U852NkNQ8

External Tools

SuperThings uses external MCP tools for research and actions:

Firecrawl - Quick Research / URL Scraping

When to use: URL tasks, single-page reads, video metadata MCP Function: mcp__firecrawl-mcp__firecrawl_scrape Best for: URLs in tasks, YouTube videos, GitHub repos, articles

Tavily - Deep Research

When to use: "Research X" tasks, DD (Deep Dive), competitive analysis MCP Function: mcp__tavily__tavily_search Best for: Multi-source research, company analysis, current events

BrowserBase - Browser Automation

When to use: Login-required sites, JavaScript-heavy pages, screenshots MCP Functions: mcp__browserbase__* Best for: Authenticated sites, form filling, dynamic pages

Zapier Gmail - Email Operations

When to use: Email tasks, intro tasks, delegation MCP Functions: mcp__zapier__gmail_* Best for: Email tasks, intro emails, delegation

Tool Selection

Task TypeTool
URL in taskFirecrawl
"Research X"Tavily
Login-required siteBrowserBase
"Email X"Zapier

Research Cache System

GTD research results cached to ~/claudesync/servers/super-things/data/research-cache.json.

  • Results persist for 7 days
  • Use /gtd resume to continue a previous session
  • Say "details N" to expand cached findings

GTD Triage Commands

CommandAction
CComplete task
C [note]Complete with note
D [person]Delegate via email
DDDeep Dive - more research
DD [focus]Deep Dive with focus

Slash Commands

  • /thingsinbox - Triage inbox with learning-based suggestions and AskUserQuestion for low confidence
  • /gtd - Get Things Done workflow with research caching

スコア

総合スコア

50/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

0/10
説明文

100文字以上の説明がある

0/10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

+5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です