WeiCHENandCursor
868238137c
Add golden-v1 hard-negative suite, shadow observability fields, query/command negation gates for high-risk writes, AllowAffirmative enforcement, candidate conflict clarification, and coordinator clarify-before-ambiguous ordering. Co-authored-by: Cursor <cursoragent@cursor.com>
150 lines
4.1 KiB
JSON
150 lines
4.1 KiB
JSON
{
|
|
"version": "intent-golden-v1",
|
|
"description": "P0 hard-negative / collision golden set for intent recognition Phase 0 baseline and Phase 1 safety gates.",
|
|
"cases": [
|
|
{
|
|
"id": "hn-claim-query",
|
|
"input": "这份报告谁认领了？",
|
|
"snapshot": {
|
|
"user_doc_role": "viewer",
|
|
"doc_lifecycle_state": "draft"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "DOCGOV_CLAIM",
|
|
"branch": "NEW_INTENT_PREEMPT",
|
|
"requires_clarifying": false
|
|
},
|
|
"expect_phase1": {
|
|
"must_not_write": true,
|
|
"dialogue_act": "ask",
|
|
"forbidden_intents": ["DOCGOV_CLAIM"]
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-submit-query",
|
|
"input": "送审通过了吗？",
|
|
"snapshot": {
|
|
"user_doc_role": "author",
|
|
"doc_lifecycle_state": "draft"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "DOCGOV_SUBMIT_REVIEW",
|
|
"branch": "NEW_INTENT_PREEMPT",
|
|
"requires_clarifying": false
|
|
},
|
|
"expect_phase1": {
|
|
"must_not_write": true,
|
|
"dialogue_act": "ask",
|
|
"forbidden_intents": ["DOCGOV_SUBMIT_REVIEW", "DOCGOV_REVIEW_APPROVE"]
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-negation-query-config",
|
|
"input": "不要修改配置，只查询当前配置",
|
|
"snapshot": null,
|
|
"expect_phase0": {
|
|
"legacy_intent": "SECURITY_FORBIDDEN",
|
|
"branch": "NEW_INTENT_PREEMPT"
|
|
},
|
|
"expect_phase1": {
|
|
"must_not_intent": "SECURITY_FORBIDDEN",
|
|
"dialogue_act": "ask"
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-affirm-plus-reg",
|
|
"input": "好，请先查法规",
|
|
"pending": {
|
|
"expected_intent": "ADVANCE_STAGE",
|
|
"allow_affirmative": true,
|
|
"default_option": {
|
|
"option_id": "1",
|
|
"title": "推进到 V0.5",
|
|
"intent": "ADVANCE_STAGE",
|
|
"slots": {
|
|
"target_stage": "V0.5_feasibility"
|
|
}
|
|
}
|
|
},
|
|
"snapshot": {
|
|
"current_stage": "V0.1_triage"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "ADVANCE_STAGE",
|
|
"branch": "EXPECTATION_HIT"
|
|
},
|
|
"expect_phase1": {
|
|
"must_not_branch": "EXPECTATION_HIT",
|
|
"must_not_intent": "ADVANCE_STAGE",
|
|
"preferred_intent": "QUERY_REGULATIONS"
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-reject-missing-reason",
|
|
"input": "把报告审核驳回",
|
|
"snapshot": {
|
|
"user_doc_role": "reviewer",
|
|
"doc_lifecycle_state": "reviewing",
|
|
"active_doc_id": "doc-1"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "DOCGOV_REVIEW_REJECT",
|
|
"branch": "AMBIGUOUS_FALLBACK",
|
|
"requires_clarifying": true
|
|
},
|
|
"expect_phase1": {
|
|
"legacy_intent": "DOCGOV_REVIEW_REJECT",
|
|
"branch": "NEW_INTENT_PREEMPT",
|
|
"requires_clarifying": true,
|
|
"clarification_contains": "驳回理由"
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-permission-mismatch",
|
|
"input": "终审批准这份文档",
|
|
"snapshot": {
|
|
"user_doc_role": "author",
|
|
"doc_lifecycle_state": "final_review",
|
|
"active_doc_id": "doc-1"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "DOCGOV_FINAL_APPROVE",
|
|
"branch": "AMBIGUOUS_FALLBACK",
|
|
"requires_clarifying": true
|
|
},
|
|
"expect_phase1": {
|
|
"legacy_intent": "DOCGOV_FINAL_APPROVE",
|
|
"branch": "NEW_INTENT_PREEMPT",
|
|
"must_not_branch": "AMBIGUOUS_FALLBACK",
|
|
"reasoning_contains": "权限不符"
|
|
}
|
|
},
|
|
{
|
|
"id": "hn-allow-affirmative-false",
|
|
"input": "好",
|
|
"pending": {
|
|
"expected_intent": "ADVANCE_STAGE",
|
|
"allow_affirmative": false,
|
|
"default_option": {
|
|
"option_id": "1",
|
|
"title": "推进到 V0.5",
|
|
"intent": "ADVANCE_STAGE",
|
|
"slots": {
|
|
"target_stage": "V0.5_feasibility"
|
|
}
|
|
}
|
|
},
|
|
"snapshot": {
|
|
"current_stage": "V0.1_triage"
|
|
},
|
|
"expect_phase0": {
|
|
"legacy_intent": "ADVANCE_STAGE",
|
|
"branch": "EXPECTATION_HIT"
|
|
},
|
|
"expect_phase1": {
|
|
"must_not_branch": "EXPECTATION_HIT"
|
|
}
|
|
}
|
|
]
|
|
}
|