Codex Skill Router v0.1.0-rc.1

Release Notes Draft: 0.1.0-rc.1

Date: 2026-06-29

This is a release candidate, not a final stable release.

What This Version Solves

0.1.0-rc.1 makes Codex Skill Router usable as a local CLI for checking whether Skills are discoverable, well-described, and likely to route as expected.

Main Commands

• csr scan: discover local Skills.
• csr audit: check Skill metadata quality.
• csr route: predict likely Skills for a task.
• csr eval: run routing regression tests.
• csr budget: estimate local Skill metadata budget.

Installation

Local development:

npm ci
npm test

Local command simulation:

npm link
csr --help

Quick Start

csr scan --path ./examples/skills
csr audit --path ./examples/skills
csr route "optimize existing page and check mobile display" --path ./examples/skills
csr eval ./examples/eval.yml --path ./examples/skills
csr budget --path ./examples/skills

Local-First and Privacy

The tool does not upload Skill contents, task prompts, Eval data, local paths, or project code. Paths are hidden by default.

Known Limitations

• Local prediction is not Codex actual Skill invocation.
• Routing is rule-based and has limited semantic understanding.
• Dependency declarations are not dependency availability checks.
• budget is a rough estimate, not real token accounting.
• The tool does not modify user Skills.
• The tool does not call external AI services.

Reporting Routing Errors

Use the GitHub "Routing error" issue template. Include:

• command;
• task prompt;
• expected include/optional/exclude;
• actual recommendations;
• whether strict behavior is expected.

Remove private paths, secrets, and private Skill contents before submitting.

Why This Is an RC

This version needs real user testing across different local Skill layouts before a stable v0.1.0 release.

What Users Should Test

• csr scan on real Skill directories;
• csr audit suggestions;
• csr route recommendations for real tasks;
• csr eval using a small custom Eval file;
• csr budget output on a real Skill set;
• path hiding behavior before sharing output publicly.