The Agent Skills Directory

[SAFE]: The skill implements a mandatory human-in-the-loop (HITL) confirmation step. It builds a triage plan and explicitly waits for the user to reply with apply before calling update_issue. This prevents unauthorized or accidental bulk modifications.
[SAFE]: Hardcoded scope restrictions (Hard Rule 1) ensure the skill only operates on the organizationSlug=sentry and project javascript. It explicitly stops and asks for confirmation if a different project is targeted.
[SAFE]: The skill follows the principle of least privilege regarding issue mutations. It is restricted to status=ignored with a specific ignoreMode=untilEscalating, preventing permanent deletion or resolution of issues without proper tracking.
[SAFE]: Data ingestion from Sentry issues (titles, stack frames) is handled through a structured classification taxonomy defined in references/archive-criteria.md. While this data is untrusted, the risk of indirect prompt injection is mitigated by the required user review of the final triage plan and the factual nature of the required reason field.

triage-frontend-issues