feat(.cursor/agents): adversarial code-review orchestrator

[tonyketcham]

Summary of changes

Adds .cursor/agents/flatbread-code-review-orchestrator.md — a system prompt that drives an adversarial code-review DAG over a Flatbread change diff using @flatbread/proof.

How it works:

• Diff intake & chunking. Materializes the diff against origin/main and builds a chunk index (path:start-end per hunk) so all findings are addressable to specific lines, not just files.
• Dynamic perspective selection (≤5). Walks the changed-file list against an 11-perspective catalog with trigger globs and selects up to five highest-impact reviewers for this diff. correctness-and-contracts and test-coverage-robustness are mandatory whenever any non-test source changes (the user's named priorities).
• Single-rank parallel review. All chosen perspectives sit in rank 1 with depends_on: []; the runner topo-sorts them into one parallel batch. Reviewers are read-only so same-rank file-write safety is satisfied.
• Rank-2 judge. Depends on every chosen perspective and merges their ## Chunks flagged / ## Coverage gaps outputs into chunk-bound structured feedback: per-chunk severity, consensus vs. disputed flagging, a consolidated ## Coverage plan (positive / negative / edge:), a verdict (APPROVE | COMMENT | REQUEST_CHANGES | BLOCK), and a reviewer scoreboard that downgrades reviewers whose findings were purely pedantic.
• Anti-pedantry guardrails. Explicit drop-list (formatting, naming, "could be more DRY") vs. keep-list (contracts, coverage, determinism, error UX) baked into both the per-perspective template and the judge's hard rules.
• Composable. Carries the ## Output Schema For DAG Handoff convention so the orchestrator itself can nest inside a parent DAG (e.g. a release-gate flow that reviews before publish).

Catalog of perspectives (orchestrator picks ≤5):

id	trigger
correctness-and-contracts	packages/*/src/**/*.ts (mandatory if matched)
test-coverage-robustness	any non-test source change (mandatory)
codegen-and-generated-artifacts	packages/codegen/**, generated TS/GraphQL
cli-and-runtime	packages/flatbread/src/cli/** + server
plugin-contract	packages/transformer-*/**, packages/source-*/**
config-and-public-api	packages/config/**
dx-and-examples	examples/**, READMEs
docs-and-positioning	*.md, docs/**
release-discipline	version bumps, scripts/publish.ts, changelog
perf-and-caching	cache/hash/dependencyCheck + hot-path resolvers
proof-runtime-internals	packages/proof/src/**

Closes #

Please don't delete this checklist! Before submitting the PR, please make sure you do the following:

• I added doc comments to any new public exports, and inline comments to any hard-to-understand areas
• My changes generate no new console errors locally
• If applicable, try to include a test that fails without this PR but passes with it — N/A; this PR is a Markdown system-prompt artifact under .cursor/agents/, not executable code. Verification is pnpm lint (prettier passes) and structural cross-checks against existing agent files (flatbread-adversarial-reviewer.md, flatbread-architecture-planner.md) plus the proof skill (.cursor/skills/proof/SKILL.md).

Does this introduce any non-backwards compatible changes?

• Yes
• No

Does this include any user config changes?

• Yes
• No