Home Explore Help
Sign In
logic855
/
claude-mods
mirror of https://github.com/0xDarkMatter/claude-mods.git
1
0
Fork 0
Files Issues 0 Wiki
Tree: 4fe9e03927
Branches Tags
claude-mods
/
commands
/
pulse.md

pulse.md 11 KB
History Raw

description: "Generate Claude Code ecosystem news digest. Fetches blogs, repos, and community sources via Firecrawl. Output to news/{date}_pulse.md."

Pulse - Claude Code News Feed

Fetch and summarize the latest developments in the Claude Code ecosystem.

What This Command Does

  1. Fetches content from blogs, official repos, and community sources
  2. Deduplicates against previously seen URLs (stored in news/state.json)
  3. Summarizes each item with an engaging precis + relevance assessment
  4. Writes digest to news/{YYYY-MM-DD}_pulse.md
  5. Updates state to prevent duplicate entries in future runs

Arguments

  • --force - Regenerate digest even if today's already exists
  • --days N - Look back N days instead of 1 (default: 1)
  • --dry-run - Show sources that would be fetched without actually fetching

Sources

Official (Priority: Critical)

[
  {"name": "Anthropic Engineering", "url": "https://www.anthropic.com/engineering", "type": "blog"},
  {"name": "Claude Blog", "url": "https://claude.com/blog", "type": "blog"},
  {"name": "Claude Code Docs", "url": "https://code.claude.com", "type": "docs"},
  {"name": "anthropics/claude-code", "url": "https://github.com/anthropics/claude-code", "type": "repo"},
  {"name": "anthropics/skills", "url": "https://github.com/anthropics/skills", "type": "repo"},
  {"name": "anthropics/claude-code-action", "url": "https://github.com/anthropics/claude-code-action", "type": "repo"},
  {"name": "anthropics/claude-agent-sdk-demos", "url": "https://github.com/anthropics/claude-agent-sdk-demos", "type": "repo"}
]

Community Blogs (Priority: High)

[
  {"name": "Simon Willison", "url": "https://simonwillison.net", "type": "blog"},
  {"name": "Every", "url": "https://every.to", "type": "blog"},
  {"name": "SSHH Blog", "url": "https://blog.sshh.io", "type": "blog"},
  {"name": "Lee Han Chung", "url": "https://leehanchung.github.io", "type": "blog"},
  {"name": "Nick Nisi", "url": "https://nicknisi.com", "type": "blog"},
  {"name": "HumanLayer", "url": "https://www.humanlayer.dev/blog", "type": "blog"},
  {"name": "Chris Dzombak", "url": "https://www.dzombak.com/blog", "type": "blog"},
  {"name": "GitButler", "url": "https://blog.gitbutler.com", "type": "blog"},
  {"name": "Docker Blog", "url": "https://www.docker.com/blog", "type": "blog"},
  {"name": "Nx Blog", "url": "https://nx.dev/blog", "type": "blog"},
  {"name": "Yee Fei Ooi", "url": "https://medium.com/@ooi_yee_fei", "type": "blog"}
]

Community Indexes (Priority: Medium)

[
  {"name": "Awesome Claude Skills", "url": "https://github.com/travisvn/awesome-claude-skills", "type": "repo"},
  {"name": "Awesome Claude Code", "url": "https://github.com/hesreallyhim/awesome-claude-code", "type": "repo"},
  {"name": "Awesome Claude", "url": "https://github.com/alvinunreal/awesome-claude", "type": "repo"},
  {"name": "SkillsMP", "url": "https://skillsmp.com", "type": "marketplace"},
  {"name": "Awesome Claude AI", "url": "https://awesomeclaude.ai", "type": "directory"}
]

Tools (Priority: Medium)

[
  {"name": "Worktree", "url": "https://github.com/agenttools/worktree", "type": "repo"}
]

GitHub Search Queries (Priority: High)

Use gh search repos and gh search code for discovery:

# Repos with recent Claude Code activity (last 7 days)
gh search repos "claude code" --pushed=">$(date -d '7 days ago' +%Y-%m-%d)" --sort=updated --limit=10

# Hooks and skills hotspots
gh search repos "claude code hooks" --pushed=">$(date -d '7 days ago' +%Y-%m-%d)" --sort=updated
gh search repos "claude code skills" --pushed=">$(date -d '7 days ago' +%Y-%m-%d)" --sort=updated
gh search repos "CLAUDE.md agent" --pushed=">$(date -d '7 days ago' +%Y-%m-%d)" --sort=updated

# Topic-based discovery (often better signal)
gh search repos --topic=claude-code --sort=updated --limit=10
gh search repos --topic=model-context-protocol --sort=updated --limit=10

# Code search for specific patterns
gh search code "PreToolUse" --language=json --limit=5
gh search code "PostToolUse" --language=json --limit=5

Reddit Search (Priority: Medium)

Use Firecrawl or web search for Reddit threads:

site:reddit.com/r/ClaudeAI "Claude Code" (hooks OR skills OR worktree OR tmux)

Official Docs Search (Priority: High)

Check for documentation updates:

site:code.claude.com hooks
site:code.claude.com skills
site:code.claude.com github-actions
site:code.claude.com mcp

Execution Steps

Step 1: Check State

Read pulse/state.json to get:

  • last_run timestamp
  • seen_urls array for deduplication
  • seen_commits object for repo tracking

If today's digest exists AND --force not specified:

Pulse digest already exists for today: news/{date}_pulse.md
Use --force to regenerate.

Fetch Script

Run the parallel fetcher:

python pulse/fetch.py --sources all --max-workers 15 --output pulse/fetch_cache.json

Step 2: Fetch Sources

For Blogs - Use Firecrawl to fetch and extract recent articles:

# Fetch blog content via firecrawl
firecrawl https://simonwillison.net --format markdown

Look for articles with dates in the last N days (default 1).

For GitHub Repos - Use gh CLI:

# Get latest release
gh api repos/anthropics/claude-code/releases/latest --jq '.tag_name, .published_at, .body'

# Get recent commits
gh api repos/anthropics/claude-code/commits --jq '.[:5] | .[] | {sha: .sha[:7], message: .commit.message, date: .commit.author.date}'

# Get recent discussions (if enabled)
gh api repos/anthropics/claude-code/discussions --jq '.[:3]'

For Marketplaces/Directories - Use Firecrawl:

firecrawl https://skillsmp.com --format markdown
firecrawl https://awesomeclaude.ai --format markdown

Step 3: Filter & Deduplicate

For each item found:

  1. Check if URL is in seen_urls - skip if yes
  2. Check if date is within lookback window - skip if older
  3. For repos, check if commit SHA matches seen_commits[repo] - skip if same

Step 4: Generate Summaries

For each new item, generate:

Precis (2-3 sentences):

Engaging summary that captures the key points and why someone would want to read this.

Relevance (1 sentence):

How this specifically relates to or could improve our claude-mods project.

Use this prompt pattern for each item:

Article: [title]
URL: [url]
Content: [extracted content]

Generate:
1. A 2-3 sentence engaging summary (precis)
2. A 1-sentence assessment of relevance to "claude-mods" (a collection of Claude Code extensions including agents, skills, and commands)

Format:
PRECIS: [summary]
RELEVANCE: [assessment]

Step 5: Write Digest

IMPORTANT: Read pulse/BRAND_VOICE.md before writing. Follow the voice guidelines.

Create news/{YYYY-MM-DD}_pulse.md with format:

# Pulse · {date in words}

{Opening paragraph: Set the scene. What's the throughline this week? What should readers care about? Write conversationally, as if explaining to a smart friend.}

---

## The Signal

{1-3 most important/newsworthy items. These get:}
- 2-paragraph summaries (150-200 words)
- Extended "Pulse insights:" (2-3 sentences)
- Source name linked to parent site

### [{title}]({url})

**[{source_name}]({source_parent_url})** · {date}

{Paragraph 1: Hook + context. Start with something interesting—a question, surprising fact, or reframing.}

{Paragraph 2: Substance + implications. What's actually in it and why it matters beyond the obvious.}

**Pulse insights:** {Opinionated take on relevance to Claude Code practitioners. Be direct, take a stance.}

---

## Official Updates

{Other items from Anthropic sources}

### [{title}]({url})

**[{source_name}]({source_parent_url})** · {date}

{1 paragraph summary (60-100 words). Hook + substance + implication in flowing narrative.}

**Pulse insights:** {1-2 sentences. Practical, specific.}

---

## GitHub Discoveries

{New repos from topic/keyword searches}

### [{repo_name}]({url})

**{author}** · {one-line description}

{1 paragraph on what it does and why it's interesting.}

**Pulse insights:** {1-2 sentences.}

---

## Community Radar

{Notable community sources, blogs, discussions}

### [{source_name}]({url}) — {pithy tagline}

{2-3 sentences on what makes this source valuable.}

---

## Quick Hits

- **[{title}]({url})**: {one-line description}
- ...

---

## The Hit List

1. **{Action}** — {Why}
2. ...

---

*{Randomised footer from BRAND_VOICE.md} · {date in words} · {suffix}*

Step 6: Update State

Update pulse/state.json:

{
  "version": "1.0",
  "last_run": "{ISO timestamp}",
  "seen_urls": [
    "...existing...",
    "...new urls from this run..."
  ],
  "seen_commits": {
    "anthropics/claude-code": "{latest_sha}",
    "anthropics/skills": "{latest_sha}",
    ...
  }
}

Keep only last 30 days of URLs to prevent unbounded growth.

Step 7: Display Summary

Pulse: 2025-12-12

Fetched 23 sources
Found 8 new items (15 deduplicated)

Critical:
  - anthropics/claude-code v1.2.0 released

Digest written to: news/2025-12-12_pulse.md

Fetching Strategy

Priority Order:

  1. Try WebFetch first (fastest, built-in)
  2. If 403/blocked/JS-heavy, use Firecrawl
  3. For GitHub repos, always use gh CLI

Parallel Fetching:

  • Fetch multiple sources simultaneously
  • Use retry with exponential backoff (2s, 4s, 8s, 16s)
  • Report progress: [====------] 12/23 sources

Error Handling:

  • If source fails after 4 retries, log and continue
  • Include failed sources in digest footer
  • Don't fail entire run for single source failure

Output Example

See news/2025-12-12_pulse.md for a complete example in the current format.

Key elements:

  • Opening 2 paragraphs set the scene conversationally (see BRAND_VOICE.md)
  • "The Signal" section gets 2 paragraphs + extended insights
  • All source names link to parent sites
  • "Pulse insights:" replaces "Why it matters"
  • "The Hit List" for actionable items (not homework—marching orders)
  • Randomised footer from BRAND_VOICE.md variations

Edge Cases

No New Items

Pulse: 2025-12-12

Fetched 23 sources
No new items found (all 12 items already seen)

Last digest: news/2025-12-11_pulse.md

Source Failures

Pulse: 2025-12-12

Fetched 23 sources (2 failed)
Found 6 new items

Failed sources:
  - skillsmp.com (timeout after 4 retries)
  - every.to (403 Forbidden)

Digest written to: news/2025-12-12_pulse.md

First Run (No State)

Initialize state.json with empty arrays/objects before proceeding.

Integration

The /pulse command is standalone but integrates with:

  • claude-architect agent - Reviews digests for actionable insights (configured in agent's startup)
  • news/state.json - Persistent deduplication state
  • Firecrawl - Primary fetching mechanism for blocked/JS sites
  • gh CLI - GitHub API access for repo updates

Notes

  • Run manually when you want ecosystem updates: Pulse
  • Use --force to regenerate today's digest with fresh data
  • Use --days 7 for weekly catchup after vacation
  • Digests are git-trackable for historical reference
  • Always read pulse/BRAND_VOICE.md before writing summaries