← ClaudeAtlas

skill-creatorlisted

Create skills that pass skill-review's five gates by construction — intent capture, prior-art scan, draft against gates, dogfood self-review. Use when authoring a new skill, refactoring an existing one, or adopting a Mall unit into this brain.
fabioc-aloha/Alex_ACT_Edition · ★ 0 · AI & Automation · score 75
Install: claude install-skill fabioc-aloha/Alex_ACT_Edition
<!-- intentional divergence from Supervisor: Edition replaces Supervisor-only path (`docs/references/skill-spec-industry-baseline.md`) with inline summary, and "adopted as Supervisor's" with "adopted by ACT". Same five gates. Audited 2026-05-31. --> # Skill Creator Author skills that pass [`skill-review`](../skill-review/SKILL.md)'s five gates by construction. The five gates are the quality bar; invert them, and you build to standard the first time. This skill teaches the spec-aligned shape — agentskills.io / Anthropic / Microsoft converge on `name` + `description` as the only required fields, with ACT discipline fields layered alongside. ## Three-level loading (the cost model) Anthropic's architectural framing, adopted by ACT: | Level | When loaded | Token cost | Content | |---|---|---|---| | **L1 — Metadata** | Always (startup) | ~100 tok/skill | Frontmatter `name`, `description`. Pre-loaded into system prompt for discovery. | | **L2 — Body** | When triggered | <5k tok | SKILL.md body. Read when `description` matches the request. | | **L3 — Bundled** | On demand | Effectively unlimited | `references/`, `scripts/`, `assets/`, `examples/`. Zero context cost until accessed. Scripts emit output, not code. | Implication: every line in SKILL.md competes with conversation context once the skill fires. Bundled files are nearly free. **When in doubt, bundle.** ## When to Use - Authoring a new skill in `.github/skills/` - Refactoring an existing skill (scope shift, split, me