Changelog

All notable changes to this plugin are listed here, newest first. To cut a release, add your notes under "Unreleased", then run python scripts/bump_version.py patch (or minor or major). It updates the plugin version and moves those notes under the new version.

[Unreleased]

[0.4.0] - 2026-10-26

• The validator now checks the Codex manifests (plugin.json, .agents/plugins/marketplace.json) against the Claude ones.

• Rewrote the six skill descriptions with the phrases people use, so they trigger more reliably.

• Added triggering tests (evals/triggers.jsonl): every skill has an obvious, a paraphrase and an unrelated query, checked by the validator and exported as tool_used graders.

• The always-on rules now default to auto, so they load only in projects that match the markers.

• Moved Hard rules to the top of every skill, above the gate.

[0.3.1] - 2026-10-23

• Added the rule to trust the user's description of their own system: check the real files, report facts only, and never argue then concede. It is in the always-on rules, the router hard rules, AGENTS.md and the agent profiles.

[0.3.0] - unreleased

• Initial skills.