v1.2.0 — Meaning-preserving plain English

Plain English now preserves technical meaning

Version 1.2 is a behavior change, not a larger list of banned words.

Earlier versions treated plain English mainly as jargon replacement. That could
silently change the claim: serialization became JSON, deployment became
production, and idempotency became merely "safe to run twice." Version 1.2
explains the behavior that the evidence actually supports.

What is different

• Meaning before brevity. Facts, certainty, obligations, conditions, scope,
  exact names, errors, and validation limits survive simplification.
• Reader-aware language. Experts keep useful domain terms. Mixed and
  nontechnical audiences get a short definition when it helps them act.
• Context-aware translation. The skill explains what this code does instead
  of applying a universal jargon dictionary.
• Tighter activation. The skill targets user-facing explanations,
  summaries, handoffs, status, and error reports. It no longer claims every raw
  log, specification, or expert message.
• Purpose-built answer shapes. Summaries, errors, walkthroughs, and handoffs
  lead with the information each reader needs.
• Regression protection. Ten adversarial cases cover uncertainty,
  must-versus-should, staging-versus-live, unknown formats, exact errors, expert
  terminology, and incomplete validation.

Package improvements

• Cursor, Claude Code, Codex, and Grok copies now come from one canonical skill.
• scripts/sync-skills.sh --check detects platform drift without rewriting
  files.
• GitHub Actions checks the manifest, skill frontmatter, platform copies,
  repository structure, and behavior-contract fixture.

Upgrade

Reinstall or update from the repository, then restart the agent so it reloads
the skill. Existing invocation phrases still work.

See CHANGELOG.md
for the complete list.