Branch: refs/heads/add-xwiki-review-skill Home: https://github.com/xwiki/xwiki-dev-llm Commit: 0efb6794e02338a37cbd335d3ad76c0a47a2e70f https://github.com/xwiki/xwiki-dev-llm/commit/0efb6794e02338a37cbd335d3ad76c... Author: Vincent Massol <[email protected]> Date: 2026-08-12 (Wed, 12 Aug 2026) Changed paths: M .claude-plugin/marketplace.json M README.md M kimi.plugin.json M opencode.jsonc M xwiki/.claude-plugin/plugin.json M xwiki/okf/index.md A xwiki/skills/xwiki-review/SKILL.md Log Message: ----------- [Misc] Add the xwiki-review skill for multi-angle change review * Add `xwiki/skills/xwiki-review/SKILL.md`, a review procedure that fans out one specialist reviewer per angle — conventions, architecture, backward compatibility, defensive conventions, performance, tests, accessibility, i18n/UX, documentation, data & migration, and conformance to the JIRA issue — over a PR, a commit range or the working tree. * Ground every angle in an OKF file or a skill, so a finding must cite a rule the project actually holds rather than generic best-practice folklore. * Gate the output: route by changed-file profile, challenge each finding with an independent skeptic that must open the cited rule and confirm it says what was claimed, drop everything below 80 confidence, and cap the report. * Keep the defensive-conventions angle to mechanical invariants (escaper present, rights check present, CSRF token present, named query parameters) and forbid security vocabulary in findings, reasoning and logs alike, per `okf/processes/security-policy.md` — public CI logs are public, so the angle reports a missing mechanism and never an attack. Hunting novel vulnerabilities is explicitly out of scope. * Register the skill in `README.md` and `xwiki/okf/index.md`, and bump the plugin version to 1.0.18 across the Claude, Kimi and opencode manifests. Co-Authored-By: Claude Opus 5 (1M context) <[email protected]> To unsubscribe from these emails, change your notification settings at https://github.com/xwiki/xwiki-dev-llm/settings/notifications