whats-new

/whats-new

Surface what's new in AI software engineering -- tool releases, product launches, emerging trends, and noteworthy voices.

Default behavior: Returns cached daily report (auto-generated at 6AM PST via Convex cron). This is instant and free.

With --force: Dispatches parallel workers to scan Reddit, Hacker News, GitHub, Dev.to, Lobsters for fresh data. Uses scripts/fetch_trends.py for direct API access.

QUICK REFERENCE

SYNTAX:
  /whats-new                          # Return cached daily report (instant)
  /whats-new --force                  # Force fresh scan (expensive)
  /whats-new --days 3                 # Custom lookback (requires --force)
  /whats-new --focus tools            # Filter: tools | releases | trends | people
  /whats-new --discovery-only         # Skip known sources, find NEW things only
  /whats-new --list                   # List recent report history

FLOW:    Check cached report → If today's exists, return it → Else run parallel workers
AUTO:    Daily cron at 6AM PST generates fresh report automatically
TRACKS:  Reddit Pulse | HN Signal | Releases & Launches | Discovery (DevTo/Lobsters/GitHub)
OUTPUT:  Prioritized briefing with TL;DR, releases, discoveries, trends

WHEN TO USE

• Daily or ad-hoc "what did I miss?" briefing
• Before starting a new project (what tools exist now?)
• Tracking competitive landscape shifts
• Finding new tools/communities to subscribe to

Use /subscribe check instead for: Just checking your existing sources Use /deep-research instead for: Deep dive into a single tool or topic Use /research instead for: Specific factual questions

---

PROGRESSIVE DISCLOSURE

Core: This file (execution algorithm, track definitions) Load at runtime: docs/sources.md (people, hashtags, subreddits, companies, discovery keywords)

---

EXECUTION ALGORITHM

[1] PARSE INPUT & DETERMINE SCOPE
    • EXTRACT flags:
      --force           Force fresh scan (skip cached report)
      --list            List recent report history
      --days N          Lookback window (default: 1, min: 1, max: 90) - requires --force
      --focus <area>    Filter: "tools" | "releases" | "trends" | "people" | "all" (default)
      --discovery-only  Skip known sources, only search for NEW things
      --skip-confirm    Execute immediately without plan approval
    • LOAD: docs/sources.md for curated source lists
    • COMPUTE: date_range = "{N} days ago" to "now"
    • COMPUTE: current_year, current_month from system date

[2] CHECK CACHED REPORT (MCP Tool - Primary Path)
    **This is the default behavior - instant and free**

    • IF --list flag:
      ├─ CALL: mcp__holocron__listWhatsNewReportsTool({ limit: 10 })
      ├─ DISPLAY: Table of recent reports with dates, finding counts
      └─ STOP (unless user asks for specific report)

    • CALL: mcp__holocron__getWhatsNewReportTool({})
      Returns: { content, generatedAt, isFromToday, stats, report }

    • IF report exists AND NOT --force:
      ├─ IF isFromToday OR report < 24 hours old:
      │   ├─ DISPLAY: Full cached report content
      │   ├─ SHOW: "Report from {generatedAt} | {findingsCount} findings"
      │   └─ STOP (cached report delivered)
      ├─ ELSE (report exists but older):
      │   ├─ SHOW: "Latest report from {date} ({N} days ago)"
      │   ├─ DISPLAY: Summary from report.summaryJson (TL;DR, top discoveries)
      │   └─ ASK: "Run fresh scan? (use --force to skip this prompt)"
      └─ IF user says "yes" → continue to [3]

    • IF --force OR no cached report:
      └─ CONTINUE to step [3] for fresh generation

[3] QUERY SUBSCRIPTION DATA (Known Ecosystem via Convex)
    Skip if --discovery-only
    • CONVEX QUERY via whatsNew.getRecentSubscriptionContent:
      Fetches subscriptionContent from the past N days,
      grouped by sourceType (youtube, newsletter, changelog, reddit, ebay)
      Returns: total count, grouped data, all items with source info
    • GROUP BY source_type: reddit, youtube, newsletter, changelog
    • EXTRACT: titles, URLs, key themes
    • This becomes "known ecosystem context" passed to workers

[4] BUILD TRACK PROMPTS
    Load sources from docs/sources.md. Build 6 track prompts:

    SCRIPT: scripts/fetch_trends.py (direct API access, no MCP needed)
    • Located at: skills/whats-new/scripts/fetch_trends.py
    • Invoke via: Bash(python3 skills/whats-new/scripts/fetch_trends.py ...)
    • Output: JSON to stdout, logs to stderr
    • Platforms: reddit, hn, bluesky, github, devto, lobsters, all

    TRACK 1: REDDIT PULSE
    Focus: What's the AI dev community buzzing about?
    • PRIMARY: fetch_trends.py reddit (direct Reddit JSON API)
      ```bash
      python3 scripts/whats-new/fetch_trends.py reddit \
        --subreddits ClaudeAI,LocalLLaMA,ChatGPT,MachineLearning,devtools,OpenAI,CursorAI,SideProject \
        --days {days} --limit 15
      ```
    • SUPPLEMENT (web search for broader signal):
      - "site:reddit.com AI coding tool {current_month} {current_year}"
    • Filter: upvotes > 10, within date_range
    • Return: Top 15 posts with title, subreddit, score, URL, 1-line summary

    TRACK 2: HACKER NEWS + BLUESKY SIGNAL
    Focus: What are builders shipping and discussing?
    • HN (via fetch_trends.py):
      ```bash
      python3 scripts/whats-new/fetch_trends.py hn --tags show_hn --min-points 20 --days {days} --limit 15
      python3 scripts/whats-new/fetch_trends.py hn --tags story --min-points 50 --days {days} --query "AI coding" --limit 10
      ```
    • Bluesky (via fetch_trends.py, requires BLUESKY_HANDLE + BLUESKY_APP_PASSWORD):
      ```bash
      python3 scripts/whats-new/fetch_trends.py bluesky --query "AI coding tool" --days {days} --limit 15
      python3 scripts/whats-new/fetch_trends.py bluesky --query "MCP server" --days {days} --limit 10
      ```
    • SUPPLEMENT: Twitter/X via exa-mcp-server x-search or web search for site:x.com
      - Key accounts from docs/sources.md (tool builders, company accounts)
      - Primary hashtags: #AItools, #DevTools, #CodingAgent, #AIcoding
      - Keywords: "just shipped", "announcing", "new release", "open sourced"
    • Return: Top 15 posts/threads with author, content summary, URL

    TRACK 3: RELEASES & LAUNCHES
    Focus: What shipped this period?
    • Check known changelogs: claude-code, cursor, openai, github-copilot, convex
    • GitHub velocity (via fetch_trends.py):
      ```bash
      python3 scripts/whats-new/fetch_trends.py github --query "topic:ai-coding" --min-stars 50 --days {days} --limit 10
      python3 scripts/whats-new/fetch_trends.py github --query "AI developer tool" --min-stars 100 --days {days} --limit 10
      ```
    • Search queries (web search):
      - "AI developer tool release {current_month} {current_year}"
      - "new coding assistant launch {current_year}"
      - "AI IDE update changelog"
    • Return: Top 10 releases with product, version, key features, URL

    TRACK 4: DISCOVERY (New & Unknown)
    Focus: What tools and trends am I NOT already tracking?
    • Dev.to (via fetch_trends.py):
      ```bash
      python3 scripts/whats-new/fetch_trends.py devto --tags ai,devtools,llm,machinelearning --days {days} --limit 15
      ```
    • Lobsters (via fetch_trends.py):
      ```bash
      python3 scripts/whats-new/fetch_trends.py lobsters --days {days} --limit 15
      ```
    • Web search queries designed to surface unknowns:
      - "new AI coding tool {current_year}"
      - "alternative to cursor" / "alternative to copilot"
      - "AI developer productivity tool"
      - "open source AI IDE"
      - "MCP server new" (Model Context Protocol ecosystem)
    • EXCLUDE: Tools already in Tier 1/2 companies (from docs/sources.md)
    • Return: Top 10 discoveries with name, description, URL, why interesting

    IF --focus flag set:
      • "tools" → Run Track 3 + Track 4 only
      • "releases" → Run Track 3 only
      • "trends" → Run Track 1 + Track 2 only
      • "people" → Run Track 2 with people emphasis

[5] PRESENT PLAN (unless --skip-confirm)
    ```
    ## What's New Scan Plan