agentleFS
davekilleendavekilleen/Dex/.agents/skills/dex-whats-new/SKILL.md

dex-whats-new

Skill509 starsChanged 3 months agoMITView on GitHub

Show recent system improvements — captured learnings plus new Claude capabilities. Use when the user says 'what's new', 'any updates to how Dex works'. Not for previewing and applying a version update; use `dex-update`. Not for unused existing features; use `dex-level-up`.

The file

Show its rules and text

We split this skill into the rules it gives an agent

21 rules. About 0 are common, 9 are distinctive, and 12 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 12

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check for improvements to your Dex system from TWO sources:

    line 8
  2. Very specific:

    Read files in System/Session_Learnings/ from last 30 days

    line 36
  3. Very specific:

    Read 06-Resources/Learnings/Mistake_Patterns.md - check for new patterns

    line 37
  4. Very specific:

    Read 06-Resources/Learnings/Working_Preferences.md - check for trends

    line 38
  5. Very specific:

    If file missing or last_check is null, treat as first run.

    line 100
  6. Very specific:

    Use WebSearch to find the latest Claude Code changelog:

    line 104
  7. Distinctive:

    Search: "Claude Code changelog 2026" or "Anthropic Claude Code releases"

    line 105
  8. Distinctive:

    When running without flags, show BOTH sections:

    line 179
  9. Very specific:

    CRITICAL: After presenting findings, you MUST update System/claude-code-state.json using the Write tool.

    line 209
  10. Very specific:

    Read current state file to preserve existing features_seen array

    line 212
  11. Distinctive:

    Add any newly discovered features to the array (avoid duplicates)

    line 213
  12. Very specific:

    Update last_check to today's date (YYYY-MM-DD format)

    line 214
  13. Very specific:

    Update last_version_seen if a new version was found

    line 215
  14. Very specific:

    Write the updated JSON back to System/claude-code-state.json

    line 216
  15. Distinctive:

    Verification: After writing, confirm to user: "State file updated - next check will compare against today's findings."

    line 235
  16. Distinctive:

    After presenting findings to the user, automatically create or enrich backlog ideas:

    line 241
  17. Very specific:

    When --full is provided, include educational deep-dives:

    line 277
  18. Distinctive:

    When evaluating relevance, categorize features:

    line 307
  19. Distinctive:

    Keep explanations in plain English, not developer jargon

    line 402
  20. Distinctive:

    Update the state file after every check

    line 404
  21. Very specific:

    Call track_event with event_name whats_new_viewed and properties:

    line 426

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- Generated from `.claude/skills/dex-whats-new/SKILL.md` by `scripts/generate-agents-skills.py`. Do not edit. -->

Check for improvements to your Dex system from TWO sources:

  1. Learnings from your usage - Mistakes, patterns, opportunities captured during sessions
  2. New Claude Code capabilities - Features that could enhance your workflows

Usage

/dex-whats-new              # Review learnings + check for Claude updates
/dex-whats-new --full       # Include detailed explanations
/dex-whats-new --learnings  # Only review session learnings
/dex-whats-new --claude     # Only check Claude Code updates

Arguments

$MODE: Optional flags

  • --full - Detailed explanations of each feature
  • --learnings - Only review learnings (skip Claude check)
  • --claude - Only check Claude updates (skip learnings)

Process

Step 0: Review Session Learnings (Unless --claude flag)

Check for learnings to review:

  1. Read files in System/Session_Learnings/ from last 30 days
  2. Read 06-Resources/Learnings/Mistake_Patterns.md - check for new patterns
  3. Read 06-Resources/Learnings/Working_Preferences.md - check for trends

Extract improvement opportunities:

  • Recurring mistakes → suggest preventive measures
  • Gaps in documentation → suggest additions
  • Workflow inefficiencies → suggest automations
  • User patterns → suggest customizations

Present findings:

🧠 LEARNINGS FROM YOUR USAGE

Since last review: [X] session learnings captured

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📊 PATTERNS IDENTIFIED

1. [Pattern name]
   Observed: [frequency] times
   Impact: [what this affects]
   Suggestion: [concrete improvement]
   
2. [Pattern name]
   ...

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 SUGGESTED IMPROVEMENTS

Based on what you've learned, here's how to improve Dex:

1. [Improvement name]
   Problem: [What's not working smoothly]
   Solution: [Specific change to make]
   Files: [What to update]
   Effort: Low/Medium/High
   
2. [Improvement name]
   ...

Want me to implement any of these? (Enter number)

If no learnings: Skip this section or show: "No session learnings captured yet. The system will learn as you use it."


Step 1: Read Current State (Unless --learnings flag)

Load System/claude-code-state.json:

{
  "last_check": "2026-01-15",
  "last_version_seen": "1.0.28",
  "features_seen": ["hooks", "sub-agents", "skills", "commands", "mcp"]
}

If file missing or last_check is null, treat as first run.

Step 2: Fetch Current Changelog

Use WebSearch to find the latest Claude Code changelog:

  • Search: "Claude Code changelog 2026" or "Anthropic Claude Code releases"
  • Primary source: Anthropic's official documentation
  • Fallback: GitHub releases, official blog posts

Focus on:

  • New features and capabilities
  • Breaking changes
  • Deprecations
  • Performance improvements

Step 3: Compare and Surface Changes

Identify what's new since last_check:

For each new feature:

  1. What it does (plain English, 1-2 sentences)
  2. Why it matters for PKM users
  3. How you could use it in Dex (concrete example)
  4. Effort to adopt (Low/Medium/High)

Step 4: Present Claude Findings

If Claude Code updates found:

🆕 CLAUDE CODE UPDATES

Last checked: [date] (X days ago)
Current version: [version]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🆕 NEW FEATURES

1. [Feature Name]
   What: [Plain English description]
   For you: [How this could improve Dex]
   Effort: Low

2. [Feature Name]
   What: [Description]
   For you: [Specific improvement idea]
   Effort: Medium

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 SUGGESTED IMPROVEMENTS

Based on what's new, here are concrete things you could add to Dex:

1. [Improvement name]
   Uses: [Which new feature]
   What it does: [Specific description]
   Pillar: [Which pillar it supports]

Want me to implement any of these? (Enter number)
Or run `/dex-improve` to workshop custom ideas.

If no Claude updates:

✅ Claude Code is up to date!

Last checked: Today
Current version: [version]

No new features since your last check.

Combined View (Default)

When running without flags, show BOTH sections:

🔄 DEX SYSTEM IMPROVEMENT REVIEW

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🧠 FROM YOUR USAGE
[Session learnings section]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🆕 FROM CLAUDE CODE
[Claude updates section]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 COMBINED IMPROVEMENTS

Pulling together learnings + new capabilities:

1. [Improvement combining both]
   Uses: [Your pattern] + [New Claude feature]
   Impact: [What this unlocks]
   
Want me to implement any of these? (Enter number or 'all')

Step 5: Update State

CRITICAL: After presenting findings, you MUST update System/claude-code-state.json using the Write tool.

Steps:

  1. Read current state file to preserve existing features_seen array
  2. Add any newly discovered features to the array (avoid duplicates)
  3. Update last_check to today's date (YYYY-MM-DD format)
  4. Update last_version_seen if a new version was found
  5. Write the updated JSON back to System/claude-code-state.json

Example update:

{
  "last_check": "2026-01-28",
  "last_version_seen": "1.0.32",
  "features_seen": [
    "hooks",
    "sub-agents",
    "skills", 
    "commands",
    "mcp",
    "parallel-sub-agents",
    "async-hooks"
  ]
}

Verification: After writing, confirm to user: "State file updated - next check will compare against today's findings."


Step 6: Backlog Synthesis (NEW — Innovation Concierge)

After presenting findings to the user, automatically create or enrich backlog ideas:

  1. Call synthesize_changelog() from Improvements MCP
    • This scans the changelog for Dex-relevant features
    • Creates new AI-authored ideas for novel capabilities
    • Enriches existing ideas with "Why Now?" evidence when a platform feature strengthens them
  2. Call synthesize_learnings() from Improvements MCP
    • Scans pending session learnings for improvement opportunities
    • Creates ideas from learnings that have concrete "suggested fix" entries
    • Enriches existing ideas when learnings relate to known backlog items
  3. Report synthesis results to user:
🤖 BACKLOG SYNTHESIS

Changelog: Scanned X entries → Created Y new ideas, enriched Z existing
Learnings: Scanned X entries → Created Y new ideas, enriched Z existing

Top ideas created/enriched:
1. [idea-XXX] Title (action: created/enriched)
2. [idea-XXX] Title (action: created/enriched)
3. [idea-XXX] Title (action: created/enriched)

Run `/dex-backlog` to see full ranked backlog.

This step is what connects external intelligence to your improvement backlog. Without it, findings are presented and forgotten. With it, every relevant discovery becomes a tracked, ranked improvement opportunity.


Full Mode (--full)

When --full is provided, include educational deep-dives:

For each feature, add:

📚 DEEP DIVE: [Feature Name]

**What it is:**
[2-3 paragraph explanation of the capability]

**How it works:**
[Technical explanation with examples]

**Real-world example:**
[Concrete scenario showing the feature in action]

**In Dex, you could:**
- [Specific application 1]
- [Specific application 2]

**To implement:**
1. [Step 1]
2. [Step 2]
3. [Step 3]

Feature Categories

When evaluating relevance, categorize features:

CategoryRelevance to DexExamples
AutomationHighHooks, triggers, scheduled tasks
PerformanceMediumFaster models, caching
ContextHighMemory, skills, knowledge bases
IntegrationHighMCP improvements, new protocols
UI/UXLowIDE features, visual changes
DeveloperLowAPI changes, SDK updates

Focus on High relevance categories. Mention Medium. Skip Low unless asked.


Capability Reference

Current Claude Code features to track:

FeatureWhat It DoesDex Potential
CommandsUser-triggered workflows/plan, /daily-review, etc.
SkillsReusable behaviors, always loadedPerson lookup, writing style
HooksAuto-triggers at eventsSession start, file changes
Sub-agentsParallel workers, isolated contextResearch, analysis
MCPExternal service connectionsCalendar, tasks, email
MemoryCross-session persistencePreferences, learnings
ToolsBuilt-in capabilitiesFile ops, search, terminal

Example Output

📢 Claude Code Updates

Last checked: 2026-01-15 (8 days ago)
Current version: 1.0.32

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🆕 NEW FEATURES

1. Async Hooks
   What: Hooks can now run without blocking the main conversation
   For you: Faster session starts — load context in background
   Effort: Low

2. Sub-agent Communication
   What: Sub-agents can now pass data back to parent
   For you: Research agent could update your notes directly
   Effort: Medium

3. MCP Resource Subscriptions
   What: MCP servers can push updates, not just respond to queries
   For you: Get notified when calendar changes, tasks update
   Effort: High (requires MCP server changes)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 SUGGESTED IMPROVEMENTS

1. Background Context Loading
   Uses: Async Hooks
   What: Load your learnings and today's plan while you type
   Pillar: Productivity

2. Auto-Research on New Topics
   Uses: Sub-agent Communication
   What: When you mention a new company, auto-research in background
   Pillar: [context-dependent]

Want me to implement any of these? (Enter 1 or 2)

Error Handling

If web search fails:

"Couldn't fetch the latest changelog. Here's what I know was current as of [last_check]: [List known capabilities]

Try again later, or check manually at docs.anthropic.com"

If state file is corrupted:

"Your capability state file has an issue. I'll create a fresh one and do a full scan."

Then proceed with first-run behavior.


Behaviors

Always Do

  • Keep explanations in plain English, not developer jargon
  • Tie every feature back to concrete Dex improvements
  • Update the state file after every check
  • Offer to implement suggestions

Never Do

  • List features without explaining relevance
  • Skip the state update
  • Overwhelm with every minor change (focus on impactful features)
  • Assume user knows Claude Code internals

Related Commands

  • /dex-improve — Full design partner (includes this + workshopping + audit)
  • /create-mcp — Build new integrations when new MCP features enable them

Track Usage (Silent)

Call the mark_feature_used tool on the dex-analytics MCP server with dex-whats-new. This ticks the feature's box in System/usage_log.md, which is what /dex-level-up reads to recommend features that have not been tried. It writes locally and sends nothing, so it runs whether or not analytics is enabled.

Analytics (Silent):

Call track_event with event_name whats_new_viewed and properties:

  • update_available (boolean)

This only fires if the user has opted into analytics. No action needed if it returns "analytics_disabled".

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.