.coderabbit.yaml

# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
#
# CodeRabbit configuration for clawbolt.
# Reference: https://docs.coderabbit.ai/reference/yaml-template
# Schema:    https://coderabbit.ai/integrations/schema.v2.json
#
# Style note: this repo bans em dashes everywhere (see AGENTS.md), including
# yaml comments. Use commas, semicolons, or rephrase.
#
# Philosophy: start small. AGENTS.md is the source of truth; CodeRabbit reads
# it via knowledge_base.code_guidelines below, so we do not restate its rules
# here. Pre-merge gates start at `warning` and only graduate to `error` once
# real PR traffic shows they catch real problems. Tools that CI already runs
# (ruff, ty, pytest, frontend typecheck) are left to CI to avoid duplicate
# signal.

language: en-US

tone_instructions: "Review code like a calm, friendly senior engineer. Be clear, constructive, and encouraging. Explain the reasoning behind suggestions, assume good intentions, and occasionally use light, friendly humor when appropriate."

# Beta features can shift under us; stay on stable until we have a reason.
early_access: false
enable_free_tier: true

reviews:
  profile: chill

  # Comment-only reviews; do not auto-request changes from CodeRabbit.
  request_changes_workflow: false

  high_level_summary: true
  high_level_summary_instructions: "Provide a concise, high-level summary of the code changes, focusing on the most significant improvements and potential areas for further enhancement. Use clear and concise language, and avoid overly technical jargon. We just want to know what changed, what moved, what was added, what was removed. Not technically. Then you can add technical jargons in another block if needed. Provide the benefits of the PR if you can"
  high_level_summary_placeholder: "@coderabbitai summary"
  high_level_summary_in_walkthrough: false

  auto_title_placeholder: "@coderabbitai title"
  auto_title_instructions: "Generate a short, action-oriented PR title. Start with a verb in the imperative form. Focus only on what changed, not why. Keep it concise and specific"

  review_status: true
  # Off to stop CodeRabbit from posting a separate per-commit status check.
  # The walkthrough comment on the PR is enough signal; CI already gates
  # merges via ruff, ty, pytest, and the frontend typecheck.
  commit_status: false
  # Fails the GitHub commit check when CodeRabbit cannot complete a review
  # (errors, timeouts, service unavailable). It does NOT fail based on review
  # findings; those are surfaced as comments.
  fail_commit_status: false

  collapse_walkthrough: false
  changed_files_summary: true
  # Auto mermaid on most diffs adds noise; opt back in once we miss it.
  sequence_diagrams: false
  estimate_code_review_effort: true
  assess_linked_issues: true
  related_issues: true
  related_prs: true

  suggested_labels: true
  labeling_instructions: []
  auto_apply_labels: false

  # Leave on so reviewer suggestions work as the maintainer team grows.
  suggested_reviewers: true
  auto_assign_reviewers: true

  in_progress_fortune: false
  poem: false

  enable_prompt_for_ai_agents: true

  path_filters: []

  # Intentionally empty. AGENTS.md and CLAUDE.md are loaded as
  # knowledge_base.code_guidelines below; restating rules here would create
  # two sources of truth that drift. Add entries only when CodeRabbit needs
  # signal that AGENTS.md cannot express.
  path_instructions: []

  abort_on_close: true
  disable_cache: false

  auto_review:
    enabled: true
    # On. CodeRabbit does proper housekeeping between incremental reviews
    # (does not keep previous reviews open) and self-caps at roughly 3
    # follow-up commits per PR before switching to manual review, so the
    # noise that originally motivated turning this off is bounded.
    auto_incremental_review: true
    ignore_title_keywords:
      - "Dependencies"
      - "update dependency"
      - "wip"
      - "dependabot"
    labels: []
    drafts: false
    base_branches: ["main"]
    ignore_usernames: []

  # Optional auto-commits CodeRabbit can offer at the end of a review. Each
  # one is a "click to accept" suggestion; nothing lands without approval.
  finishing_touches:
    docstrings:
      enabled: true
    unit_tests:
      enabled: true
    simplify:
      enabled: true

  # Gates that run before merge. Start everything at `warning` so we can
  # observe real PR traffic before promoting any of these to `error`.
  pre_merge_checks:
    docstrings:
      mode: warning
      threshold: 80
    issue_assessment:
      mode: warning
    title:
      mode: warning
      requirements: |
        PR titles should start with a Conventional Commit prefix: `feat:`, `fix:`, `docs:`, `refactor:`, `chore:`, `test:`, `perf:`, or `ci:`.
        Use imperative mood. Keep the title under ~70 chars and let the body carry the detail.
    description:
      mode: warning

  tools:
    # CI already runs `uv run ruff check`, `uv run ty check`, and `uv run
    # pytest`; leaving the equivalent CodeRabbit tools off avoids duplicate
    # signal in PR comments. The frontend has no eslint config (typecheck and
    # knip run in CI), so eslint stays off too.
    ruff:
      enabled: false
    eslint:
      enabled: false
    # Things CI does not cover and that are cheap to surface inline.
    gitleaks:
      enabled: true
    markdownlint:
      enabled: true
    yamllint:
      enabled: true
    github-checks:
      enabled: true
      timeout_ms: 90000
    # Off by default; opt back in once we know what kind of noise they make.
    shellcheck:
      enabled: false
    languagetool:
      enabled: false
    checkov:
      enabled: false
    actionlint:
      enabled: false
    semgrep:
      enabled: false
    dotenvLint:
      enabled: false

chat:
  art: true
  auto_reply: true

knowledge_base:
  opt_out: false
  web_search:
    enabled: true
  code_guidelines:
    enabled: true
    # AGENTS.md is the canonical rule set; CLAUDE.md is a symlink to it but
    # we list both since the symlink may not be resolved by the scanner.
    # DESIGN.md carries visual rules referenced by AGENTS.md.
    filePatterns:
      - "**/AGENTS.md"
      - "**/CLAUDE.md"
      - "**/DESIGN.md"
  # `auto` keeps learnings per-repo unless the same content shows up in a
  # sibling repo. clawbolt and clawbolt-premium share an open-core codebase
  # and conventions, so cross-repo learning is welcome. Switch to `global`
  # deliberately if the shared rule set grows.
  learnings:
    scope: auto
  issues:
    scope: auto
  pull_requests:
    scope: auto
  mcp:
    usage: enabled

code_generation:
  docstrings:
    language: en-US

# Issue enrichment intentionally disabled for now. Revisit once we have a
# clearer sense of CodeRabbit's baseline behavior on real PRs.
issue_enrichment:
  auto_enrich:
    enabled: false
  planning:
    enabled: false
    auto_planning:
      enabled: false
      labels: []