.coderabbit.yaml

# CodeRabbit config — agent-deck (public OSS Go repo)
# Docs: https://docs.coderabbit.ai/reference/configuration

language: en

reviews:
  profile: chill                    # less nitpicky, ship-friendly
  request_changes_workflow: false   # never block merge with "Request Changes"
  high_level_summary: true
  poem: false                       # save tokens; the bunny is cute but eats budget
  review_status: true
  collapse_walkthrough: true

  # Surface the PR scope + scoring up-front
  changed_files_summary: true       # file-by-file change summary
  sequence_diagrams: true           # Mermaid diagrams for complex flows (great for refactors)
  estimated_code_review_effort: true # review-time estimate badge
  assess_linked_issues: true         # check PR ↔ issue alignment
  related_issues: true               # surface related open issues
  related_prs: true                  # surface related open/recent PRs
  suggested_labels: true             # auto-label proposal
  suggested_reviewers: true          # nominate reviewers from CODEOWNERS

  # Auto-suggest follow-ups: missing docstrings, missing unit tests
  finishing_touches:
    docstrings:
      enabled: true
    unit_tests:
      enabled: true

  # SCORES + quality gates (the "score" you saw on CodeRabbit)
  # Each runs the AI quality assessor; thresholds 0-100 block merge if not met.
  # Start lenient; tighten over time as we build confidence.
  pre_merge_checks:
    docstrings:
      mode: warning                 # warn, don't block
      threshold: 70
    title:
      mode: warning
      requirements: "Conventional Commits format (feat:, fix:, chore:, etc.)"
    description:
      mode: warning
      requirements: "PR description must explain why + reference issue (Closes #N)"
    issue_assessment:
      mode: warning                 # PR must address what linked issue actually asks
      threshold: 70
    custom_checks:
      - name: "remote_parity"
        mode: warning
        instructions: |
          If this PR changes TUI behavior in internal/ui/ or cmd/agent-deck/,
          verify it also handles RemoteSession (or has an explicit t.Skip
          documenting why). See ~/.agent-deck/skills/pool/agent-deck-tdd-feature/SKILL.md.
      - name: "test_coverage_per_surface"
        mode: warning
        instructions: |
          Any new feature must have tests for every applicable surface
          (TUI / Web / CLI / remote). Stubs in Web UI are not acceptable for new features.

  auto_review:
    enabled: true
    drafts: false                   # don't review draft PRs (RFCs etc.)
    base_branches:
      - main
    ignore_title_keywords:
      - "[skip ci]"
      - "[no-review]"
      - "WIP"

  # Don't review pure dependency bumps or pure-docs churn (CI tools already cover):
  path_filters:
    - "!.github/dependabot.yml"
    - "!**/*.md"
    - "!CHANGELOG.md"
    - "!**/testdata/**"
    - "!documentation/**"           # design docs don't need code review

  # Bots that open PRs should NOT trigger reviews or consume seats:
  ignore_users:
    - dependabot[bot]
    - github-actions[bot]
    - renovate[bot]

  # Path-specific instructions — extra scrutiny on hot paths
  path_instructions:
    - path: "internal/session/**"
      instructions: |
        This is the core session lifecycle. Pay extra attention to:
        - signal handling (SIGINT, SIGTERM teardown)
        - SQLite write atomicity
        - RemoteSession vs LocalSession parity
        - race conditions (this code runs with -race in CI)
    - path: "internal/web/**"
      instructions: |
        Web backend. Verify:
        - new endpoints update tests/web/PARITY_MATRIX.md
        - mutator parity with TUI (any state change in TUI must work in Web too)
        - handlers have a corresponding _test.go
    - path: "internal/tmux/**"
      instructions: |
        tmux integration. Verify:
        - tmux command construction uses slice args (not shell strings)
        - controlling-tty assumptions documented (see #1114 for the failure mode)
    - path: "cmd/agent-deck/**"
      instructions: |
        CLI surface. Verify:
        - flags follow existing pattern (--listen, --no-tui, etc.)
        - --help output reflects the new behavior
    - path: ".github/workflows/**"
      instructions: |
        CI workflows. Verify:
        - third-party actions pinned by SHA or major-version tag
        - timeouts on every test/build step (we hit a 3hr hang earlier)
        - permissions block follows least-privilege

  tools:
    # Go-specific linters already run in CI; CodeRabbit can summarize their output
    golangci-lint:
      enabled: true
    gosec:
      enabled: true
    yamllint:
      enabled: true
    shellcheck:
      enabled: true
    actionlint:                     # GitHub Actions linting
      enabled: true
    hadolint:                       # if we ever add Dockerfiles
      enabled: true
    biome:                          # for the tests/web TypeScript
      enabled: true
    markdownlint:
      enabled: false                # noisy on docs-heavy repos

chat:
  auto_reply: true                  # respond to @coderabbitai mentions
  integrations:
    jira:
      usage: auto                   # if we ever wire jira
    linear:
      usage: auto

# Code generation suggestions (used when @coderabbitai is asked to write code)
code_generation:
  docstrings:
    language: en-US
  unit_tests:
    language: en-US

knowledge_base:
  opt_out: false                    # share learnings (OSS is public anyway)
  learnings:
    scope: auto                     # learn this repo's patterns over time
  issues:
    scope: local                    # link related issues for context
  pull_requests:
    scope: local                    # link related PRs for context