Your skills are leaving your hands. Don't let a rent-a-brain keep them.
protocolsagents
read at source ↗ natesnewsletter.substack.com
Your skills are leaving your hands. Don’t let a rent-a-brain keep them.
Source: Nate’s Newsletter Date: 2026-06-19 URL: https://natesnewsletter.substack.com/p/claude-codex-agent-skills
Summary
Nate’s Newsletter argues that agent skills — prompts, SKILL.md files, runbooks, MCP configs — are becoming trapped inside specific platforms, creating lock-in analogous to earlier memory and context lock-in. Workers who build workflows inside one tool lose them when they switch. The core prescription: own the way you work by structuring skills as portable artifacts that travel with you, not vendor-specific configuration.
Implications
- Agent layer: Skills-as-code (SKILL.md, runbooks) are emerging as a first-class portability surface; teams that don’t treat them as owned artifacts accumulate hidden switching costs.
- Floor/trust: Vendor dependency on skill execution is a trust boundary — if the platform owns the skill, it can alter behavior silently across updates.
- Capability clocks: Portability pressure across Claude Code, Codex, and Cursor signals that the agentic tooling market is still in format-war phase; no standard has won.