Source Intelligence

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

All of v2.0.20 Home All releases olderv2.0.19 v2.0.21newer
Claude Code v2.0.20

Skill Tool (Full Implementation)

What: Complete implementation of the Skill tool, enabling Claude to execute skills and commands directly during conversations (previously existed only as a stub)

How to use: Skills are invoked automatically by Claude when appropriate, or you can configure permission rules:

# In .claude/settings.local.json
{
  "permissions": {
    "allow": {
      "Skill": ["pdf", "xlsx", "custom-skill:*"]
    }
  }
}

Details:

  • Supports both mode commands (high-priority) and regular skills
  • Includes intelligent token budgeting (15,000 character default via SLASH_COMMAND_TOOL_CHAR_BUDGET)
  • Permission system with wildcard patterns (e.g., skill-name:*)
  • Skills are categorized by source: project, user, plugin, or managed
  • Mode commands receive priority and full token allocation
  • Evidence: Mo1 initialization at line 429046, W$Q() prompt generator at line 428983, nw variable at line 428903

See this entry in the whole of v2.0.20 →