A meta-skill for building Claude skills, plus the skills it produces.

This repo is both a skill and a skill registry:
- skill-forge/           the authoritative meta-skill (v1.0.0)
- modal-llm-inference/   produced-by-skill-forge reference skill (v0.3.1)
- contrib/               curated index of community skills (links, not vendored)
- scripts/list_skills.py regenerates the root README index from SKILL.md frontmatter

Each skill is installable standalone via gh skill (v2.90+):
  gh skill install curtismercier/skill-forge skill-forge
  gh skill install curtismercier/skill-forge modal-llm-inference

skill-forge ships with:
- Three-tier complexity ladder (single-file, scripts+refs, full system)
- Source-first research via gh_research.py (stdlib only)
- Four audit passes: validate, security scan, anti-slop audit, staleness
- ERRATA lifecycle for tracked drift
- Verification playbook + CLAUDE.md scaffolding for Tier 3 skills
- Behavioral-eval tooling (aggregate_benchmark, eval_viewer, grader/comparator/analyzer
  agent prompts) schema-compatible with anthropics/skills skill-creator
- Vendored + hardened execution scripts from anthropics/skills (Apache 2.0) with
  skill-forge wrapper adding preflight/error-separation/strict-length/exit-code/SIGINT
  fixes
- End-to-end walkthrough (skill-forge/docs/10-walkthrough.md) building a real Tier 2
  skill with every tool — every command runnable, every output real

Licensed under MIT (this repo's code); vendored scripts remain Apache 2.0 with
attribution preserved. See LICENSE and LICENSE-APACHE-ANTHROPIC.txt.

Compatible with Claude Code, Claude Desktop, claude.ai, Copilot, Cursor, Codex CLI,
Gemini CLI, Kiro, Antigravity via standard SKILL.md format.