CEL Triggers and Normalized Events

Harness trigger expressions — and the CEL filters in .feature fixtures that exercise them — match against the normalized event model, not raw forge-specific webhook payloads. Whenever you write or review a CEL trigger, use the normalized transition.kind vocabulary. A raw forge webhook action name used where a transition.kind is expected silently evaluates to false, so the trigger never fires and nothing warns you.

Source of truth:docs/normative/normalized-event/v1/README.md. Its Transition kind vocabulary, Transition sub-objects, and CEL trigger examples sections are authoritative for the full enumerated kind list and the sub-object each kind requires (label, comment, review, …). Do not re-copy that list here — a duplicated table drifts from the schema.

Highest-value pitfall — synchronize vs synchronized: GitHub's raw webhook action is synchronize (no trailing d), but the normalized kind is synchronized. A CEL expression written as event.transition.kind == "synchronize" silently never matches — this exact typo appeared in a fork-dispatch scenario and was eliminated in PR #5309. The same raw→normalized gap exists elsewhere (e.g. labeled / unlabeled → label_changed); when unsure, check the vocabulary section above rather than guessing from the webhook name.

When reviewing harness triggers or .feature CEL filters: flag any raw forge webhook action name used where a normalized transition.kind is expected (e.g. "synchronize", "labeled") as a medium-severity finding — it will silently fail to match.