hardening(security): release gate ignores note-level code-scanning alerts (zizmor class)
Hardening idea from the review of #6622 (no action requested there to keep the PR scoped):
The release gate's check-code-scanning-alerts job fails only on critical/high/medium code-scanning alerts. Note- and low-severity findings never fail a gate, and zizmor's SARIF upload under min-severity: low filters informational findings — so a whole class of supply-chain hygiene findings (e.g. zizmor/self-repository, which produced 40 Note-level alerts #8218–#8257 that sat open for weeks) is doubly invisible: excluded from gating and can accumulate silently on main.
Two candidate hardenings (either, not necessarily both):
- A dedicated release-gate step that queries code scanning for open alerts of selected rules (e.g.
zizmor/*) regardless of severity, failing the gate if any exist — forcing triage-or-dismiss instead of silent accumulation. - Lower the failure threshold for the zizmor scan specifically (its findings are Note-level by design; severity is not the right signal there — presence is).
Sizing note: #6622's review verified the current gap empirically (the 40 self-repository alerts never gated anything; they closed only via the next SARIF upload after conversion). A follow-up deciding "alert presence vs severity" as the zizmor gate signal would close the class permanently.
Source: LearningCircuit/local-deep-research