Codex Skill Router v0.1.0

Release Notes: 0.1.0

Date: 2026-07-02

Codex Skill Router v0.1.0 is the first stable release of the local Skill routing quality checker for Codex.

What It Does

Codex Skill Router helps users inspect, audit, route, evaluate, and estimate local Codex Skills before relying on them in real work.

It answers practical questions:

• Which Skills are discoverable?
• Are Skill descriptions clear enough?
• Which Skills look suitable for this task?
• Did a routing change improve or regress behavior?
• How large is the local Skill metadata context?

Install

After npm publication:

npm install -g codex-skill-router
csr --help

Repository install:

git clone https://github.com/zcccc-mxc/codex-skill-router.git
cd codex-skill-router
npm install
npm link
csr --help

Main Commands

csr scan
csr audit
csr route "check login authorization bypass"
csr eval ./examples/eval.yml --path ./examples/skills
csr budget

Local-First Guarantees

• No API key required.
• No external AI service required.
• No cloud database.
• No Skill upload.
• No modification of user Skill files.
• Paths are hidden by default.

RC1 Validation Result

v0.1.0-rc.1 completed real validation before this stable release decision:

30 real tasks reviewed
28 complete
2 failed
Required Skill Recall: 100.0%
No-Match Accuracy: 100.0%
Exclusion Accuracy: 91.7%
npm test: 21 tests passed

Known P2 issue:

docs-authoring may be over-recommended in a small number of web/mobile tasks.
The correct primary Skills are still selected.
This does not block v0.1.0.

Current Limitations

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

Reporting Routing Errors

When reporting a routing problem, include:

• command used;
• sanitized task prompt;
• expected include, optional, and exclude;
• actual recommendations;
• whether no-match was expected.

Remove private paths, secrets, customer names, and private Skill contents before posting.

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.