.coderabbit.yaml

# CodeRabbit review configuration.
#
# Chill rather than assertive, and it never requests changes. That is not a
# preference about tone — it is about who holds the merge decision. The
# auto-merge gate in .github/workflows/auto-merge.yml decides mechanically
# whether a PR is safe to land; a bot that could request changes would add a
# second, softer gate whose failures are advisory but whose blocking is real.
#
# CodeRabbit's job here is to give a contributor a useful read before a human
# looks, and to produce structured fix instructions that an AI agent can
# consume via @coderabbitai autofix.
language: en-US

reviews:
  profile: chill
  request_changes_workflow: false

  high_level_summary: true
  review_status: true
  poem: false

  finishing_touches:
    autofix:
      enabled: true

  auto_review:
    enabled: true
    drafts: false

  path_instructions:
    - path: "**/*.go"
      instructions: >
        This project's value is arithmetic correctness about money. Flag any
        float64 in a pricing, rate or amount path — decimal.Decimal is required
        everywhere money is handled. Flag any value that defaults to zero when
        data is unavailable: unknown must be reported as unknown, never
        defaulted, guessed or averaged away. Flag any figure presented to a
        user that did not come from a live source or a recorded snapshot.
        When you find an issue, include a Prompt for AI Agents block with
        exact fix instructions so @coderabbitai autofix can resolve it.
    - path: "**/*_test.go"
      instructions: >
        Tests must run from testdata/snapshots via snapshot.Replayer, never the
        live network. Flag any test that constructs a fixture from the package's
        own wire structs rather than recorded bytes — such a fixture can only
        prove the parser agrees with itself. Flag happy-path-only tests: a test
        that cannot fail proves nothing.
    - path: "docs/**"
      instructions: >
        Docs here are contracts. Flag any claim about a capability, figure or
        deployment that is not currently true, and any measurement quoted
        without its date and source.