AI Development Pulse

Telemetry (run first, silent):

aiframework-telemetry skill_invoked skill=aif-pulse 2>/dev/null || true

Stay current with Claude Code features, best practices, and the AI-assisted development ecosystem. Discovers what's new and applies relevant improvements to your project.

When to Use

• Weekly or after major Claude Code releases
• When you suspect there are new features you're not using
• When you want to optimize your AI development workflow
• After reading about a new Claude Code capability

Process

Step 1: Research Latest Updates

Search for the latest Claude Code documentation and features:

WebSearch: "Claude Code new features 2026"
WebSearch: "Claude Code changelog latest release"
WebSearch: "Claude Code best practices update"
WebSearch: "claude code skills hooks agents new"

Also check official sources:

• WebFetch: https://code.claude.com/docs/en/changelog (if available)
• WebFetch: https://docs.anthropic.com/en/docs/claude-code/overview

Step 2: Audit Current Project Setup

Read the project's current configuration:

1. CLAUDE.md — how many lines? Is it lean or bloated?
2. .claude/rules/ — what rules exist? Are they path-scoped?
3. .claude/skills/ — what skills exist? Are they well-structured?
4. .claude/settings.json — what permissions/hooks are configured?
5. .claude/agents/ — any custom subagents?
6. AGENTS.md — cross-tool compatibility?
7. .mcp.json — any MCP servers configured?
8. Check git hooks: .githooks/ or .husky/

Step 3: Gap Analysis

Compare current setup against latest best practices:

CLAUDE.md Health Check:

• Is it under 200 lines? (Official recommendation)
• Does it contain ONLY what Claude can't figure out from code?
• Are there rules that belong in .claude/rules/ instead?
• Are there workflows that belong in .claude/skills/ instead?
• Is there a Self-Evolution section?

Rules Check:

• Are rules using paths: frontmatter for file-scoped loading?
• Are rules specific enough? (not just generic "write clean code")
• Are there domains detected in the manifest that don't have rules?

Skills Check:

• Is there a review skill? Ship skill? Learn skill?
• Are skills using allowed-tools to scope permissions?
• Are skills using disable-model-invocation for side-effect workflows?
• Could any repeated prompts become skills?

Settings Check:

• Are safe tools pre-approved? (Read, Glob, Grep, WebSearch)
• Are there hooks for auto-formatting after edits?
• Are dangerous commands blocked? (rm -rf, DROP TABLE)

Advanced Features Check:

• Custom subagents in .claude/agents/?
• MCP servers for external tool integration?
• CLAUDE.local.md for personal overrides?
• Prompt caching optimization?

Step 4: Discover New Capabilities

Search for features the project might benefit from:

WebSearch: "Claude Code hooks PostToolUse auto format"
WebSearch: "Claude Code custom subagents .claude/agents"
WebSearch: "Claude Code MCP servers useful"
WebSearch: "Claude Code skills examples advanced"
WebSearch: "Claude Code context window optimization"

Look specifically for:

• New tool types — has Claude Code added new built-in tools?
• New hook events — FileChanged, CwdChanged, PreCompact, PostCompact?
• New skill features — context: fork, agent: Explore, model overrides?
• New settings — auto-memory, env injection, model selection?
• Community patterns — what are top projects doing with .claude/?

Step 5: Generate Recommendations

For each finding, create a recommendation:

## Recommendations

### 1. [Feature Name]
- **What**: Brief description of the feature/practice
- **Why**: How it helps this specific project
- **How**: Exact steps or code to implement
- **Impact**: High/Medium/Low

### 2. ...

Categories:

• Quick Wins — can be applied in <5 minutes
• Medium Effort — needs 15-30 minutes of setup
• Strategic — requires planning, significant benefit

Step 6: Apply Quick Wins

For recommendations marked as Quick Wins, ask the user:

"I found N improvements. M are quick wins I can apply now. Should I apply them?"

If approved, apply each one:

• Update .claude/settings.json with new hooks/permissions
• Add missing .claude/rules/ files
• Update CLAUDE.md if it's bloated
• Create new skills for repeated workflows
• Add .claude/agents/ for specialized tasks

Step 7: Log and Track

1. Record findings to the learnings file:

echo '{"date":"'$(date +%Y-%m-%d)'","category":"pattern","summary":"AI Pulse findings","detail":"Found N improvements, applied M quick wins","files":["CLAUDE.md",".claude/"]}' >> tools/learnings/*-learnings.jsonl

2. Update vault status:

| {date} | pulse | AI Pulse: found N improvements, applied M |

Append to vault/wiki/log.md

3. Update vault/memory/status.md with pulse date

What to Look For (Reference)

Claude Code Features Checklist

AI Development Patterns