No Cap

/no-cap

Extract signal from noise. Pulls your recent X/Twitter bookmarks, filters out marketing/spam/engagement bait, and extracts actionable intelligence. Dual output: structured markdown for agents, HTML email digest for humans.

Commands

Setup Flow (/no-cap setup)

Guide the user through configuration:

1. Sign into X — Detect the user's platform and guide accordingly:

macOS users — Run the auto-login command:

   cd {repoPath} && npx tsx src/cli.ts auto-login
   

Windows / Linux users — Auto-login is macOS only. Guide them through the manual method: 1. Open Chrome → go to x.com 2. Press F12 (DevTools) → Application tab → Cookies → https://x.com 3. Copy the values for auth_token and ct0 4. Run:

   cd {repoPath} && npx tsx src/cli.ts update-cookies  
   

2. Output directory — Ask where to save signals (default: ~/no-cap-signals/)

3. Interests & sections — Ask: "What topics or projects are you tracking? These become sections in your digest." - Let the user list topics freeform (e.g., "AI agents", "fine-tuning LLMs", "marketing", "Claude Code") - Also ask if they have active projects they want signals routed to (e.g., "my-app — AI dashboard", "trading bot") - Save these as interests and projects arrays in config - These guide how signals get categorized and routed — they're not hard filters, just lenses

4. Email digest — Ask if they want an HTML email digest after each run. If yes: - Ask for their email address - Ask for their Resend API key (free at https://resend.com — 100 emails/day on free tier) - No MCP plugins or heavy dependencies needed

5. Write config to ~/.no-cap/config.json using the Write tool. Set file permissions to 600 (owner-only read/write) since it contains session credentials. Make sure repoPath is set in the config to the directory where no-cap is installed.

6. Test run — Do a test fetch to verify credentials work:

   cd {repoPath} && npx tsx src/cli.ts fetch
   

Main Loop (/no-cap)

First, read the no-cap config to get the repo path:

Step 1: Fetch new bookmarks

Run the CLI to get new bookmarks:

cd {repoPath} && npx tsx src/cli.ts fetch

This returns JSON with new bookmarks (already diffed against state). If "No new bookmarks" — tell the user and stop.

Step 2: Noise Filter (3 layers)

For each bookmark, evaluate through three layers. Be aggressive — most content is noise.

Layer 1 — Account-level:

Layer 2 — Post-level:

Layer 3 — Content-level:

Important: Signal-in-the-noise. A post can be mostly noise but still contain a real idea buried inside the hype. For example, a Polymarket trading post might be 80% marketing fluff but contain a genuinely novel arbitrage mechanism. A course seller might drop a real technique in their pitch thread. When you filter a post as noise, always check: is there a kernel worth extracting? If yes, classify it as partial_signal — extract the idea but note what's noise around it. These go in a separate "Signal in the Noise" section, not the main signals.

Step 3: Classify bookmark intent

Before extracting signals, classify WHY the user bookmarked this. People bookmark for different reasons:

| Intent | Description | Example | |--------|-------------|---------| | signal | Contains an actionable insight, technique, or pattern | "MCTS for agent decision-making replaces brute-force loops" | | action_item | Something the user should DO — a course, exam, event, tool to try, tutorial to follow | "Claude Code architect certification exam is now open" | | project_signal | Relevant to a specific project the user is working on (check config projects) | A post about WASM sandboxing when user has an agent platform project | | learning | Educational content worth saving for reference — deep dives, papers, threads | "Comprehensive thread on transformer architecture internals" | | tool | A tool, library, or resource worth bookmarking | "New CLI tool for X/Twitter data export" |

Do NOT filter out action items as noise. A post about "how to pass the Claude Code architect exam" is not marketing — it's something the user wants to be reminded to do. Same for events, courses, tutorials, deadlines.

Step 4: Extract signals and action items

For each bookmark that passed noise filtering, extract based on its intent:

For signals and project_signals:

Title: One-sentence summary of the core idea
Type: technical_insight | market_signal | workflow | tool | trend | opinion
Intent: signal | project_signal
Project: [which project, if project_signal — match against config projects]
Actionability: immediate | near_term | reference
Tags: [freeform topic tags — try to match user's configured interests]
Source: @handle, URL, date
Context: Why this matters — 2-3 sentences. Strip narrative, isolate the mechanism.

For action items:

Title: What to do — imperative form ("Take Claude Code architect exam", "Try Resend for email")
Intent: action_item
Priority: high | medium | low
Source: @handle, URL, date
Context: Why this is worth doing — 1-2 sentences

For learning resources:

Title: What this teaches
Intent: learning
Tags: [topic tags]
Source: @handle, URL, date
Context: What you'll learn — 1-2 sentences

If the bookmark has a quoted tweet or linked post (from link propagation), include that context in the analysis.

Step 5: Write agent output

Session file — Write to {signalDir}/{YYYY-MM-DD}/session-{N}.md:

# YYYY-MM-DD — Session N
Signals
[Theme Name]
[Signal 1] — [2-3 sentence context with the mechanism, why it matters, and source] (@handle)
[Signal 2] — [context] (@handle, @handle2)
[Theme Name 2]
[Signal] — [context] (@handle)
Project Signals
[Only if user has projects configured]
[Project Name]:
[Signal] — [why it matters for this project] (@handle)
Action Items
[ ] [What to do] — [brief context] (@handle) — Priority: high/medium/low
Signal in the Noise
[Extracted idea] — [context]. Source is noisy because [reason]. (@handle)
Noise
[What the post was about] — Filtered because: [specific reason] (@handle)
Stats
Processed: X | Signals: Y | Actions: Z | Noise: N

Determine session number by counting existing session files in today's directory.

Master file — Read {signalDir}/MASTER.md, then rewrite to reflect cumulative state. This file is YOUR primary reference — write it so you (Claude) can load it next session and immediately understand what's happening, what matters, and what to do.

Structure:

# No Cap — Signal Master
_Last updated: YYYY-MM-DD_
_Sessions: N total | Latest: YYYY-MM-DD_
New Since Last Session
[Quick diff — what changed this run. 2-3 sentences max. Remove this section when the next session runs.]
Project Signals
[Grouped by the user's configured projects. Each signal has enough context to act on without reading the source.]
[Project Name]
[Signal title]
[2-3 sentences: what it is, why it matters for this project, what the mechanism/technique is. Include enough detail that you could reason about it or suggest actions without going back to the tweet.]
Source: @handle · YYYY-MM-DD · actionability
Connects to: [what existing project work this relates to]
Unrouted
[Signals that don't map to any configured project but are still actionable]
Action Items
[ ] [Action] — [context] (source, date) — Priority: high/medium/low
Noise Patterns
[Recurring noise patterns across sessions — accounts/topics to auto-filter. This section compounds over time.]
@handle — [pattern] (seen N times)
Historical
[Signals that have been acted on or are >30 days stale. Compressed to one-liners.]

Writing principles for MASTER.md:

Step 6: Generate and send email digest (if configured)

Generate the email digest HTML using the EXACT template structure below. Do NOT improvise the layout — use this template and fill in the content. The template has been tested across email clients and renders correctly.

How to use the template: 1. Copy the template below 2. Replace {date}, {theme_count}, {action_count} in the header 3. Replace {tldr} with 2-3 sentence TL;DR 4. For each action item: copy the action card block, fill in icon emoji, title, description, and source link 5. For each signal theme: copy the signal card block, fill in icon emoji, heading, prose paragraph, and source handles 6. For secondary/reference signals (lower priority): use the two-column card block 7. For project signals (if configured): use the project card block 8. For noise: fill in the noise entries 9. Write the completed HTML to {signalDir}/{YYYY-MM-DD}/digest-{N}.html

IMPORTANT: Use valign="top" on all icon elements. Set icon

  
  
  
  




  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  

    

      

        

          
no cap
          
Signal from noise · {date}
        
        

          
{theme_count} themes · {action_count} actions
        
      
    
  

    
{tldr}
  

    
To Do
  

    

      

        

          

            {action_icon}
          
        
        

          
{action_title}
          
{action_description} — {action_source}
        
      
    
  
  

    
Signals
  

    

      

        

          

            {signal_icon}
          
        
        

          
{signal_heading}
          
{signal_prose}
          
{signal_sources}
        
      
    
  
  

    

      

        

          

            

              

                {col1_icon}
              
              
{col1_heading}
              
{col1_prose}
              
{col1_note}
            
          
        
        

          

            

              

                {col2_icon}
              
              
{col2_heading}
              
{col2_prose}
              
{col2_note}
            
          
        
      
    
  
  

    

      

        

          

            {project_icon}
          
        
        

          
{project_name}
          
{project_signals_prose}
        
      
    
  
  

    
Filtered · {noise_count}
    
    
{noise_entry}
  

    

      

        

          
no cap
          
Signal from noise · Built with Claude Code
        
        

          
Powered by No Cap skill
        
      
    
  

cd {repoPath} && npx tsx src/cli.ts send-email "No Cap — {N} signals from {date}" "{signalDir}/{YYYY-MM-DD}/digest-{N}.html"

⚠️  FIRST-RUN NOTICE: CHECK YOUR SPAM FOLDER
Emails from No Cap are sent via Resend (from onboarding@resend.dev).
Your email provider may flag the first one as spam.
If you don't see it in your inbox, check spam/junk and mark it as "Not Spam"
so future digests land in your inbox.

cd {repoPath} && npx tsx src/cli.ts mark-done {newestBookmarkId}

No Cap — {date}
Processed: {X} bookmarks
Signals: {Y} | Action items: {Z} | Learning: {W}
Noise filtered: {N}
Saved to: {signalDir}/{date}/session-{N}.md
Master updated: {signalDir}/MASTER.md
{Email sent to: address | Email digest saved to: path}

cd {repoPath} && npx tsx src/cli.ts status

cd {repoPath} && npx tsx src/cli.ts update-cookies  

"Not logged into X" or "Missing auth_token"