docs(readme): grfp pipeline refresh — parser-warm + 6 langs + persistent HTTP lead

[kellenff]

Summary

5-stage README refresh synthesized from the GRFP (GitHub README For Perfectionists) pipeline:

• Deep Dive → understood the surface (21 tools, 6 languages, 1 dep, MCP-native)
• Crystal Ball → identified complexity hotspots, corrected the stale "Python next" claim (Python/PHP/Rust are all wired)
• Brain Jam → chorus-locked the angle (parser-warm + 6 langs + persistent HTTP), transcript at .brainstorm/chorus-20260713T184229.json
• Think Tank → extracted 6 adoptable moves from 6 exemplars (ripgrep, ast-grep, Serena, codebase-memory-mcp, Playwright MCP, Deno)
• Pen Wielding → wrote the final README

What changed in the README

• Lead reframed — parser-warm + 6 languages (Go/TS/JS/Python/PHP/Rust) + persistent HTTP, with the 200k LOC / 3-8s reparse number doing real work as a budget.
• Trust strip expanded — SLSA L3 · 21 tools · 6 languages · 1 dep · read-only. The "6 languages" addition is the previously-missing receipt for the polyglot claim.
• New ## Transports section — explicit stdio vs persistent HTTP comparison + Mermaid diagram. Persistent HTTP promoted from install sub-bullet to primary section, since it's the architectural anchor (parser state survives agent restarts).
• New ## URI handling section — explicit normalization rules so agents that build URIs (Claude/Cline/Goose normalize differently) don't trip on sloppy path handling.
• 21-tool table gained a read-only column — only 2 of 21 mutate (the registry cache operations). Playwright-MCP pattern.
• New 6-group taxonomy (introspect / traverse / search / diagnose / registry / workflow) replaces the prior 16/4/1 structural grouping.
• New ## When yactt is the wrong tool section — honest Serena contrast table + 4 explicit trade-off beats. Earns trust by naming the trade (LSP-grade fidelity vs cross-language consistency + parser-warm persistence).
• Roadmap corrected — Python/PHP/Rust moved from "next" to "shipped". New next-milestone: cross-repo graph queries.
• Receipts section expanded — 5 claims → 5 receipts.

What did NOT change

• All 21 tool names (canonical source of truth in internal/tool/register.go)
• The 5-call payment validation worked example (reconciled to real tool names; args now include file:// URIs)
• The Mermaid quadrantChart (auto-renders to SVG)
• The Trust & security section (AST02/03/05/09 mitigations)

grfp/* reports

All five stage reports are refreshed in .claude/grfp/:

• deep-dive.md — refreshed 2026-07-13 (graph grew 1,858 → 3,123 nodes)
• crystal-ball.md — refreshed with graph-confirmed hotspots
• brain-jam.md — refreshed with chorus transcript
• think-tank.md — refreshed (Deno + Playwright MCP re-read, 2 new moves)

Plus .brainstorm/chorus-20260713T184229.json — the chorus transcript.

Test plan

• Render the Mermaid quadrantChart block — should auto-render to SVG
• Render the Mermaid graph LR diagram in ## Transports
• Spot-check the 21-tool table against internal/tool/register.go
• Spot-check the worked example against a real yactt mcp serve session