# yaml-language-server: $schema=https://cubic.dev/schema/cubic-repository-config.schema.json
#
# Cubic config — make the AI reviewer use the same guidance as the repo-local
# `/cf-review` skill and human reviewers. See https://docs.cubic.dev/configure/cubic-yaml
#
# Single source of truth: the review guidance lives in skills/cf-review/SKILL.md.
# Edit the skill, not this file. (Cubic reads at most ~10k chars per custom
# agent and the skill is larger, so the durable, principle-level lenses are
# restated in `description` — concatenated first, so always kept — while
# `file_paths` carries the full detail. The rot-prone specifics, e.g. exact
# canonical-home module names, live only in the skill.)

version: 1

reviews:
  # Posture mirrors the skill's bandwidth discipline: flag real bugs and
  # regressions loudly, keep nits minimal, don't demand a pristine first pass.
  sensitivity: medium

  custom_rules:
    - name: cf-review (Common Fabric review guidance)
      description: >-
        Apply Common Fabric's code-review guidance (full text follows, sourced
        from skills/cf-review/SKILL.md). Repo-specific lenses to apply even if
        the text below is truncated: (1) Anti-duplication — never re-fork
        hashing, serialization, cloning, or identity; reuse the canonical
        @commonfabric modules instead of hand-rolling. (2) Framework-fit — in
        patterns and the runtime that serves them, flag code fighting the
        transformer pipeline / reactive-graph model (error-swallowing try/catch,
        singletons, async/await in handlers, transformer escape hatches). (3) Coherence — a change to runtime
        semantics must update the docs, comments, and examples it makes stale,
        so someone searching later is not misled. (4) Test rigor — each touched
        test should guard a nameable principle. (5) Bandwidth — separate
        blocking bugs from non-blocking improvements and optional nits.
      file_paths:
        - skills/cf-review/SKILL.md