v0.7.0 — schema evidence layer + stdlib prgenius package + stdio MCP shell

v0.7.0 — Foundation (2026-07-04)

This is the foundation release that everything since builds on.
If you are upgrading, read this first, then look at v0.7.1 for
the prgenius → prgenius-kb rename and v0.7.3 for evidence coverage.

T1 — prgenius Python package (new, src/prgenius/)

• pyproject.toml (stdlib-only, no deps)
• parser.py — pure-stdlib YAML-subset frontmatter parser (zero PyYAML)
• cli.py — subcommands: profile, case, schema, dump, mcp
• mcp.py — stdio MCP shell exposing 4 tools:
  • get_repo_profile(repo)
  • list_open_prs()
  • get_case_study(repo, pr_number)
  • schema_info()
• NDJSON dump output (benchmark-ready)
• Smoke-tested only (no formal test suite — tracked as TODO G2+)

T2 — schema v0.7.0 (BC over v0.5.0)

• ROUNDS_SCHEMA.md adds three optional fields, available at both levels:
  • Round-level: delta.verified_at, delta.evidence_urls, delta.confidence
  • Case-level: verified_at, evidence_urls, confidence
• 100% BC: when absent they're ignored; presence is forward-compatible
• validate.py rewritten to surface v0.7.0 fields and emit
  --enforce-evidence warnings when case-level evidence is missing

T3 — stdio MCP shell (scripts/dashboard.py)

• Reads data/snapshot.json when present, falls back to in-script
  computation
• --snapshot mode emits JSON snapshot of n-counts
• Stats line consumes snapshot numbers (avoids hand-written drift)

T4 — README narrative cleanup

• External-facing README + index.md use neutral terms
• Internal history (CHANGELOG / KNOWN_ISSUES) uses maintainer / initiator
• "Pragmatic" CHANGELOG lines preserved (data, not narrative)
• Dropped "MisakaNet 一侧" section (planned-only, not actionable)

T5 — legacy BC scaffold

• validate.py --enforce-evidence surfaces cases needing v0.7.0 fields
• The 100%-coverage goal from evaluation is now checkable, not aspirational

Stats

• 12 profiles / 12 case studies / 11 lessons / 5 anti-patterns
• snapshot.json: 60 files / 12 profiles / 11 case studies / 0 errors / 0 warnings
• data/ in .gitignore (don't ship generated snapshots)

Important upgrade note

In v0.7.1 the distribution was renamed prgenius → prgenius-kb to
avoid a PyPI name collision with an unrelated 2024 GPT-3 PR-description
tool. If you are installing for the first time, use prgenius-kb.