Changelog

1.2.0 — 2026-10-08

• Wired the accumulating failure log into the process: step 1 reads references/llm-failure-log.md before the checklist, step 6 appends any genuinely new pattern (one entry per new pattern, never per run, never delete an entry)
• The log existed since 1.1.0 but nothing in the process ever invoked it; this closes that loop so a caught pattern is checked first on the next review
• Register instruction no longer stalls: apply the caller's own voice skill when one is loaded, and only ask which register when there is no such skill and the user has not stated a preference
• Version bumped in both manifests (plugin.json and .claude-plugin/plugin.json)

1.1.0 — 2026-08-19

• Added register profile (references/voice-profile.md) applied after the correctness pass
• Added runnable checklist, accumulating terminology glossary, and accumulating failure log
• Added explicit "write in Arabic" trigger (اكتب بالعربية / اكتبلي بالعربية)
• Resolved dialect negation mapping (مش/موش → ليس generally, لم only for past negation)
• Trimmed description under the agentskills.io 1024-char limit; added author/license/version frontmatter
• Restructured to a polyglot layout: Agent Plugins 1.0.0 (plugin.json) + Claude Code marketplace (.claude-plugin/) + skills.sh

1.0.0 — 2026-08-19

• Initial release: QALB-grounded Arabic correctness (Sections 1-7) with QALB spelling, worked MT examples, and dialect classification references