@daviss.dev: Here's what most people miss. A skill is not a saved prompt. It's a markdown file with frontmatter that tells Claude when to load it, what tools it can use, and what to do. It reads files. Writes files. Calls APIs. Spawns subagents. Triggers other skills. Two locations matter: - ~/.claude/skills/ runs globally, follows you across every project - .claude/skills/ stays in the repo, only fires in that codebase Personal workflow goes global. Client and project work stays local. The unlock nobody talks about: skills chain. A transcript-extraction skill hands its output to a project-tracker skill, which hands its output to a follow-up draft. Three small skills, one command, thirty minutes of work gone. Go one level higher and you stop using Claude Code like a coding assistant. You map your entire job into a system: - Skills do the repeatable mechanics - Hooks fire automatically on events - Agents handle the parts that need their own reasoning That is systems engineering applied to how you actually work. The build process that actually produces skills you'll reuse: - Separate objective from subjective. Mechanics vs judgment - Ideate with Claude before writing the file - Use voice transcription to dump every detail of how the work happens - Build a meta-skill that turns those voice notes into a clean SKILL.md - Test on real messy input, not a sanitized example - Refactor after the first few uses #claudecode #ai #aiagents #aiautomation #buildinpublic

Davis McMurrain

Open In TikTok:

Region: US

Saturday 30 May 2026 21:52:38 GMT