Version 1.0.0
Included
- Five evidence-based scope classifications.
- Component-level request decomposition and overall aggregation.
- Evidence, uncertainty, and stop-condition rules.
- Structured output contract, validator, and Markdown renderer.
- Five fictional canonical scenarios and expected outputs.
- Source-agnostic operation for content the host can already read.
- Codex, Claude Code, and Claude.ai installation guidance.
Safety boundaries
- No legal advice or binding contract decisions.
- No invented price, rates, delivery dates, or schedule impact.
- No automatic messages or completed approvals.
- Low-confidence and legally disputed cases withhold client-facing drafts.
Known limitations
- Source access and extraction depend on the host.
- Python validation is optional and requires Python 3 plus PyYAML.
- Deterministic validation checks structure and internal consistency, not factual correctness.
- Compatibility must be rechecked when hosts change their skill surfaces.