Author-Model Declaration

Why Declare

• An LLM reviewing its own output misses its own blind spots and favors its own style — cross-family review catches what same-family review waves through
• The reviewer workflow can only pick a different-family model if the PR signals which model authored the code
• Missing declaration is grounds for REQUEST_CHANGES — reviewers exit without reading the diff rather than silently reviewing as human, because an unannounced AI author is the exact case the rule exists to catch

How to Declare

• Explicit (preferred): add an **Author-Model:** line to the PR body, near the top: **Author-Model:** claude-opus-4-7
• Implicit (fallback): a Co-authored-by: git trailer on any commit in the PR (Claude Code emits this by default)
• Human-authored PRs: **Author-Model:** human
• Mixed authorship: list every model that contributed, space-separated — **Author-Model:** human claude-opus-4-7

Precedence

• Every PR must carry a signal — the declaration is required, not optional
• PR body Author-Model: wins when both exist (explicit beats implicit)
• Trailer-only: the reviewer extracts the model from the trailer's display name — known names normalize to canonical IDs (e.g., Claude Opus 4.7 → claude-opus-4-7); unknown display names are still accepted as ad-hoc model IDs so a trailer is never silently rejected
• Neither present is a policy violation — the reviewer exits with REQUEST_CHANGES before reading the diff (no degraded-fallback review), so the missing declaration blocks the PR until it's added

Model Families

• claude-* → anthropic
• gpt-*, codex-* → openai
• gemini-* → google
• Unknown strings default to their own ad-hoc family — unknown never matches a known family

Mixed Authorship

• If any AI model contributed, the reviewer family must differ from every declared AI family — worst-case matching, not most-permissive
• **Author-Model:** human claude-opus-4-7 is treated as claude-authored for family-mismatch purposes; the openai-family reviewer runs, the claude-family reviewer skips
• If the declaration spans both paired reviewer families (e.g., gpt-5.4 claude-opus-4-7), no cross-family reviewer is available — both run as a degraded fallback rather than leaving the PR unreviewed. Author teams that want to preserve cross-family review should avoid co-authoring across both paired families on the same PR
• If the declaration spans neither paired reviewer family (e.g., gemini-2.5, a custom model name, or human-only), both paired reviewers run. Both ARE cross-family relative to the author, so neither is biased; the duplicate review is accepted noise. Picking one reviewer arbitrarily would silently halve coverage on PRs from other-family or human-only authors, which is worse than the duplication. The neither-family case is therefore deliberate, not an oversight in the gate logic