What problem does it solve? Golden and regression fixtures often pass today but flake tomorrow because they assert approximate output, cover only the happy path, or depend on ambient repo state. This Skill enforces a discipline for authoring committed, byte-stable fixtures that actually prove the guards and branches they claim to cover. ## Core Features & Use Cases - Byte-Stable Assertions: Pins canonical JSON key order and explicitly sorted iteration order so expected output is byte-identical across runs and machines. - Both-Branch Guard Coverage: Requires a negative/inverse fixture for every guard (stale vs. fresh, ceiling-hit vs. no-ceiling-hit, success vs. fallback), including fixtures that manufacture the actual failure condition before asserting recovery. - Composition and Coherence Fixtures: Adds cross-stage fixtures where one capability's output feeds another's input, and joint assertions when multiple products must share one identifier such as a provenance header or generation-id. - Use Case: When extending a deterministic pipeline stage (e.g., a graphify-backed context generator), author one named fixture per live consumer, per branch, and per composition point so regressions surface immediately in CI. ## Quick Start Ask the agent to author a golden fixture for the new pipeline stage, covering both the passing and failing branches of its staleness guard with byte-identical expected output.
