v0.5.3 — Trust, DX, and Brief Scoring

v0.5.3 turns the project from a protocol plus validation toolkit into a more
measurable agent workflow.

Added

• JSON brief scoring:
  • agenda-intelligence score brief.json
  • returns a heuristic 0-100 structural quality score
  • reports relevance, evidence support, completeness, actionability, and clarity
• Evidence-linked scoring:
  • agenda-intelligence score brief.json --evidence evidence-pack.json
  • accounts for supported, partially supported, and unsupported claims
  • accounts for missing and contradicting sources
  • checks evidence-mode alignment and retrieved_at hygiene for live-backed packs
• MCP read tools:
  • get_protocol
  • list_lenses
  • get_lens
  • source_plan
• Regression tests for package-data consistency, MCP read tools, JSON scoring, and
  evidence-linked scoring.

Changed

• Synchronized package version, top-level manifest, and packaged manifest to
  0.5.3.
• Synchronized top-level and packaged agenda brief schemas.
• Updated README, quickstart, evaluation docs, MCP docs, roadmap, and skill docs
  to match implemented behavior.
• Kept before/after markdown scoring working while adding JSON brief scoring.

What this does not do

• It does not verify factual truthfulness.
• It does not fetch live sources from a brief.
• It does not replace human or LLM judge review for high-stakes analysis.

The scoring layer is a practical quality loop for structure and evidence
discipline, not a truth engine.