customer-support Tools
support:triage accepts supplied ticket snapshots and approved knowledge and returns routing, escalation evidence and unsent response drafts. See the pack guide for the strict input, policy settings, errors and provenance. This page’s tables are generated by pnpm docs:generate:pack-reference; notes outside the managed block are preserved.
Tool Summary
Section titled “Tool Summary”| Tool | Permission | Required approvals | Required scopes |
|---|---|---|---|
support:triage | read | None | network off |
Tool Details
Section titled “Tool Details”support:triage
Section titled “support:triage”- Permission:
read - Required approvals: None
- Required scopes:
network off
Input schema
Section titled “Input schema”Output schema
Section titled “Output schema”Not declared in the pack manifest.
The manifest rejects invalid outer envelopes and unknown root fields before dispatch with INVALID_INPUT; its unprojected array items keep the raw tickets for digesting. The handler validates the strict ticket and knowledge shapes and returns SUPPORT_INPUT_INVALID, SUPPORT_INPUT_LIMIT, SUPPORT_REPORT_TOO_LARGE or SUPPORT_KNOWLEDGE_UNAPPROVED. Canonical policy comes from trusted host context under the customer_support key; caller input and selector metadata cannot replace it.
Success returns success: true with data: status (DRAFT), input_bytes, input_digest, ticket counts (ticket_count, routed_count, unknown_count, human_required_count, draft_count), draft_text_limit, detail_level, the actions receipt (zero sends, zero ticket mutations, zero network calls), approved knowledge citations, per-ticket tickets (including matched_escalations with each configured reason), policy_echo, effective_policy (or null with a policy_summary when large), configuration_digest and a summary. Failure returns success: false with an error code and actionable message. No response is sent, no provider ticket is changed, no URL is fetched and no model is called. Future billing remains unvalidated.