agnostic-ai

AI tooling changed again.

Release briefings that separate shipped agnostic-ai support from verified upstream CLI and model news.

Current edition
Week of 15 September 202625 targets checked

New safety boundaries are becoming project configuration

Manual-only skills, agent-scoped MCP access, directory-scoped skills, project plugins, and ten confirmed target gaps.

Read the briefing
designClaudeCrush

Manual-only skills now share one safety intent

Both tools can stop the model from invoking a sensitive skill automatically.

designQoderFactoryOpenHands

Agents can receive narrower MCP access

Three tools expose least-privilege controls through three different native shapes.

fixCodexCursor

Nested skills carry directory scope

Both adapters flatten them today, turning local guidance into repository-wide guidance.

watch5 tools

Project plugins are becoming a shared workflow

The user goal is common. Execution, package, ownership, and trust models are not.

All editions

Subscribe by RSS

New safety boundaries are becoming project configuration

Manual-only skills, agent-scoped MCP access, directory-scoped skills, project plugins, and ten confirmed target gaps. Amp Augment Claude Code Cline Codex CLI GitHub Copilot Crush Cursor Factory Junie Kiro OpenCode OpenHands Qoder Windsurf 10 findings ยท 3 decisions