ai-capability-framework 1.0.0-rc.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +37 -0
- package/CONTRIBUTING.md +56 -0
- package/LICENSE +21 -0
- package/README.md +198 -0
- package/SECURITY.md +29 -0
- package/conformance/invalid/duplicate/capabilities/conformance.duplicate.one.yaml +55 -0
- package/conformance/invalid/duplicate/capabilities/conformance.duplicate.two.yaml +55 -0
- package/conformance/invalid/eval-results/conformance.results.failing.json +15 -0
- package/conformance/invalid/missing-reference/capabilities/conformance.missing_reference.prepare.yaml +58 -0
- package/conformance/invalid/schema/capabilities/conformance.invalid.missing_required.yaml +3 -0
- package/conformance/valid/capabilities/conformance.note.get.yaml +82 -0
- package/conformance/valid/decisions/conformance.note.get.denied_missing_permission.json +13 -0
- package/conformance/valid/entities/Note.yaml +18 -0
- package/conformance/valid/eval-results/conformance.results.passing.json +19 -0
- package/conformance/valid/evals/conformance.note.get.valid.yaml +33 -0
- package/conformance/valid/openai/context.no_permissions.json +6 -0
- package/dist/adapter-common.d.ts +62 -0
- package/dist/adapter-common.d.ts.map +1 -0
- package/dist/adapter-common.js +453 -0
- package/dist/ai-sdk.d.ts +5 -0
- package/dist/ai-sdk.d.ts.map +1 -0
- package/dist/ai-sdk.js +71 -0
- package/dist/anthropic-claude.d.ts +5 -0
- package/dist/anthropic-claude.d.ts.map +1 -0
- package/dist/anthropic-claude.js +62 -0
- package/dist/cli.d.ts +11 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +360 -0
- package/dist/decision.d.ts +6 -0
- package/dist/decision.d.ts.map +1 -0
- package/dist/decision.js +237 -0
- package/dist/eval-runner.d.ts +6 -0
- package/dist/eval-runner.d.ts.map +1 -0
- package/dist/eval-runner.js +372 -0
- package/dist/gemini.d.ts +5 -0
- package/dist/gemini.d.ts.map +1 -0
- package/dist/gemini.js +64 -0
- package/dist/generated/manifest-types.d.ts +208 -0
- package/dist/generated/manifest-types.d.ts.map +1 -0
- package/dist/generated/manifest-types.js +1 -0
- package/dist/index.d.ts +15 -0
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +13 -0
- package/dist/langchain.d.ts +5 -0
- package/dist/langchain.d.ts.map +1 -0
- package/dist/langchain.js +65 -0
- package/dist/loader.d.ts +4 -0
- package/dist/loader.d.ts.map +1 -0
- package/dist/loader.js +94 -0
- package/dist/mcp.d.ts +5 -0
- package/dist/mcp.d.ts.map +1 -0
- package/dist/mcp.js +82 -0
- package/dist/openai-responses.d.ts +6 -0
- package/dist/openai-responses.d.ts.map +1 -0
- package/dist/openai-responses.js +65 -0
- package/dist/registry.d.ts +6 -0
- package/dist/registry.d.ts.map +1 -0
- package/dist/registry.js +134 -0
- package/dist/semantic-kernel.d.ts +5 -0
- package/dist/semantic-kernel.d.ts.map +1 -0
- package/dist/semantic-kernel.js +79 -0
- package/dist/types.d.ts +400 -0
- package/dist/types.d.ts.map +1 -0
- package/dist/types.js +1 -0
- package/dist/validator.d.ts +3 -0
- package/dist/validator.d.ts.map +1 -0
- package/dist/validator.js +105 -0
- package/docs/adapter-roadmap.md +179 -0
- package/docs/adapters.md +112 -0
- package/docs/api.md +115 -0
- package/docs/capability-manifests.md +47 -0
- package/docs/control-plane.md +59 -0
- package/docs/eval-runner.md +80 -0
- package/docs/evals.md +34 -0
- package/docs/getting-started.md +51 -0
- package/docs/host-responsibilities.md +23 -0
- package/docs/interoperability.md +32 -0
- package/docs/migration-0.1-to-1.0.md +22 -0
- package/docs/openai-responses.md +103 -0
- package/docs/release.md +75 -0
- package/docs/spec.md +194 -0
- package/examples/eval-results/public.results.passing.json +128 -0
- package/examples/scheduling/capabilities/scheduling.availability.get.yaml +91 -0
- package/examples/scheduling/capabilities/scheduling.invite.prepare.yaml +114 -0
- package/examples/scheduling/capabilities/scheduling.invite.send.yaml +97 -0
- package/examples/scheduling/decisions/scheduling.availability.get.select.json +14 -0
- package/examples/scheduling/decisions/scheduling.invite.prepare.allowed.json +20 -0
- package/examples/scheduling/decisions/scheduling.invite.send.allowed.json +22 -0
- package/examples/scheduling/decisions/scheduling.invite.send.denied_missing_approval.json +18 -0
- package/examples/scheduling/entities/AvailabilitySlot.yaml +26 -0
- package/examples/scheduling/entities/MeetingInvite.yaml +25 -0
- package/examples/scheduling/eval-results/scheduling.results.passing.json +59 -0
- package/examples/scheduling/evals/scheduling.availability.get.valid.yaml +38 -0
- package/examples/scheduling/evals/scheduling.invite.prepare.valid.yaml +44 -0
- package/examples/scheduling/evals/scheduling.invite.send.requires_approval.yaml +43 -0
- package/examples/scheduling/openai/context.scheduler.json +6 -0
- package/examples/support/capabilities/support.refund.commit_case.yaml +98 -0
- package/examples/support/capabilities/support.refund.prepare_case.yaml +116 -0
- package/examples/support/capabilities/support.ticket.get.yaml +85 -0
- package/examples/support/decisions/support.refund.commit_case.allowed.json +22 -0
- package/examples/support/decisions/support.refund.commit_case.denied_missing_approval.json +18 -0
- package/examples/support/decisions/support.refund.prepare_case.approval_required.json +19 -0
- package/examples/support/decisions/support.ticket.get.select.json +0 -0
- package/examples/support/entities/Order.yaml +23 -0
- package/examples/support/entities/Ticket.yaml +25 -0
- package/examples/support/eval-results/support.results.passing.json +68 -0
- package/examples/support/evals/support.refund.commit_case.requires_approval.yaml +42 -0
- package/examples/support/evals/support.refund.prepare_case.approval_required.yaml +45 -0
- package/examples/support/evals/support.refund.prepare_case.valid.yaml +43 -0
- package/examples/support/openai/context.support_agent.json +9 -0
- package/package.json +74 -0
- package/schemas/capability-manifest.schema.json +250 -0
- package/schemas/entity-manifest.schema.json +97 -0
- package/schemas/eval-case.schema.json +134 -0
- package/schemas/eval-result.schema.json +77 -0
|
@@ -0,0 +1,250 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://raw.githubusercontent.com/xennon101/ai-capability-framework/main/schemas/capability-manifest.schema.json",
|
|
4
|
+
"title": "AICF Capability Manifest",
|
|
5
|
+
"type": "object",
|
|
6
|
+
"additionalProperties": false,
|
|
7
|
+
"required": [
|
|
8
|
+
"schema_version",
|
|
9
|
+
"id",
|
|
10
|
+
"version",
|
|
11
|
+
"status",
|
|
12
|
+
"name",
|
|
13
|
+
"summary",
|
|
14
|
+
"model_description",
|
|
15
|
+
"capability_type",
|
|
16
|
+
"autonomy_tier",
|
|
17
|
+
"risk_tier",
|
|
18
|
+
"input_schema",
|
|
19
|
+
"output_schema",
|
|
20
|
+
"side_effects",
|
|
21
|
+
"authorization",
|
|
22
|
+
"policy",
|
|
23
|
+
"lifecycle",
|
|
24
|
+
"observability"
|
|
25
|
+
],
|
|
26
|
+
"properties": {
|
|
27
|
+
"schema_version": {
|
|
28
|
+
"const": "1.0"
|
|
29
|
+
},
|
|
30
|
+
"id": {
|
|
31
|
+
"type": "string",
|
|
32
|
+
"pattern": "^[a-z][a-z0-9_]*(\\.[a-z][a-z0-9_]*){2,}$"
|
|
33
|
+
},
|
|
34
|
+
"version": {
|
|
35
|
+
"type": "string",
|
|
36
|
+
"pattern": "^(0|[1-9]\\d*)\\.(0|[1-9]\\d*)\\.(0|[1-9]\\d*)(?:-[0-9A-Za-z-.]+)?(?:\\+[0-9A-Za-z-.]+)?$"
|
|
37
|
+
},
|
|
38
|
+
"status": {
|
|
39
|
+
"enum": ["draft", "experimental", "active", "deprecated", "disabled"]
|
|
40
|
+
},
|
|
41
|
+
"name": {
|
|
42
|
+
"type": "string",
|
|
43
|
+
"minLength": 1
|
|
44
|
+
},
|
|
45
|
+
"summary": {
|
|
46
|
+
"type": "string",
|
|
47
|
+
"minLength": 1
|
|
48
|
+
},
|
|
49
|
+
"model_description": {
|
|
50
|
+
"type": "string",
|
|
51
|
+
"minLength": 1
|
|
52
|
+
},
|
|
53
|
+
"domain": {
|
|
54
|
+
"type": "string",
|
|
55
|
+
"minLength": 1
|
|
56
|
+
},
|
|
57
|
+
"owner": {
|
|
58
|
+
"type": "object",
|
|
59
|
+
"additionalProperties": false,
|
|
60
|
+
"properties": {
|
|
61
|
+
"team": { "type": "string" },
|
|
62
|
+
"contact": { "type": "string" }
|
|
63
|
+
}
|
|
64
|
+
},
|
|
65
|
+
"capability_type": {
|
|
66
|
+
"enum": [
|
|
67
|
+
"read_data",
|
|
68
|
+
"retrieve_documents",
|
|
69
|
+
"compute",
|
|
70
|
+
"write_prepare_only",
|
|
71
|
+
"write_commit",
|
|
72
|
+
"external_message_prepare",
|
|
73
|
+
"external_message_send",
|
|
74
|
+
"workflow_start",
|
|
75
|
+
"workflow_step",
|
|
76
|
+
"human_handoff"
|
|
77
|
+
]
|
|
78
|
+
},
|
|
79
|
+
"autonomy_tier": {
|
|
80
|
+
"enum": ["A0", "A1", "A2", "A3", "A4", "A5"]
|
|
81
|
+
},
|
|
82
|
+
"risk_tier": {
|
|
83
|
+
"enum": ["none", "low", "medium", "high", "critical"]
|
|
84
|
+
},
|
|
85
|
+
"when_to_use": {
|
|
86
|
+
"type": "array",
|
|
87
|
+
"items": { "type": "string", "minLength": 1 }
|
|
88
|
+
},
|
|
89
|
+
"when_not_to_use": {
|
|
90
|
+
"type": "array",
|
|
91
|
+
"items": { "type": "string", "minLength": 1 }
|
|
92
|
+
},
|
|
93
|
+
"input_schema": { "$ref": "#/$defs/json_schema_object" },
|
|
94
|
+
"output_schema": { "$ref": "#/$defs/json_schema_object" },
|
|
95
|
+
"side_effects": {
|
|
96
|
+
"type": "object",
|
|
97
|
+
"additionalProperties": false,
|
|
98
|
+
"required": [
|
|
99
|
+
"reads_data",
|
|
100
|
+
"writes_data",
|
|
101
|
+
"creates_records",
|
|
102
|
+
"updates_records",
|
|
103
|
+
"deletes_records",
|
|
104
|
+
"sends_external_messages",
|
|
105
|
+
"charges_money",
|
|
106
|
+
"refunds_money",
|
|
107
|
+
"changes_permissions",
|
|
108
|
+
"triggers_external_workflow",
|
|
109
|
+
"irreversible"
|
|
110
|
+
],
|
|
111
|
+
"properties": {
|
|
112
|
+
"reads_data": { "type": "boolean" },
|
|
113
|
+
"writes_data": { "type": "boolean" },
|
|
114
|
+
"creates_records": { "type": "boolean" },
|
|
115
|
+
"updates_records": { "type": "boolean" },
|
|
116
|
+
"deletes_records": { "type": "boolean" },
|
|
117
|
+
"sends_external_messages": { "type": "boolean" },
|
|
118
|
+
"charges_money": { "type": "boolean" },
|
|
119
|
+
"refunds_money": { "type": "boolean" },
|
|
120
|
+
"changes_permissions": { "type": "boolean" },
|
|
121
|
+
"triggers_external_workflow": { "type": "boolean" },
|
|
122
|
+
"irreversible": { "type": "boolean" }
|
|
123
|
+
}
|
|
124
|
+
},
|
|
125
|
+
"authorization": {
|
|
126
|
+
"type": "object",
|
|
127
|
+
"additionalProperties": false,
|
|
128
|
+
"required": ["permissions", "tenant_scoped", "requires_user_context"],
|
|
129
|
+
"properties": {
|
|
130
|
+
"permissions": {
|
|
131
|
+
"type": "array",
|
|
132
|
+
"minItems": 1,
|
|
133
|
+
"items": { "type": "string", "minLength": 1 }
|
|
134
|
+
},
|
|
135
|
+
"tenant_scoped": { "type": "boolean" },
|
|
136
|
+
"requires_user_context": { "type": "boolean" },
|
|
137
|
+
"data_scope": {
|
|
138
|
+
"type": "array",
|
|
139
|
+
"items": { "type": "string", "minLength": 1 }
|
|
140
|
+
}
|
|
141
|
+
}
|
|
142
|
+
},
|
|
143
|
+
"policy": {
|
|
144
|
+
"type": "object",
|
|
145
|
+
"additionalProperties": false,
|
|
146
|
+
"minProperties": 1,
|
|
147
|
+
"properties": {
|
|
148
|
+
"approval_required": { "type": "boolean" },
|
|
149
|
+
"approval_required_if": {
|
|
150
|
+
"type": "array",
|
|
151
|
+
"items": { "$ref": "#/$defs/policy_rule" }
|
|
152
|
+
},
|
|
153
|
+
"deny_if": {
|
|
154
|
+
"type": "array",
|
|
155
|
+
"items": { "$ref": "#/$defs/policy_rule" }
|
|
156
|
+
},
|
|
157
|
+
"max_autonomy_tier": {
|
|
158
|
+
"enum": ["A0", "A1", "A2", "A3", "A4", "A5"]
|
|
159
|
+
}
|
|
160
|
+
}
|
|
161
|
+
},
|
|
162
|
+
"lifecycle": {
|
|
163
|
+
"type": "object",
|
|
164
|
+
"additionalProperties": false,
|
|
165
|
+
"required": ["prepare", "preview", "approve", "commit", "verify", "audit"],
|
|
166
|
+
"properties": {
|
|
167
|
+
"prepare": { "type": "boolean" },
|
|
168
|
+
"preview": { "type": "boolean" },
|
|
169
|
+
"approve": { "type": "boolean" },
|
|
170
|
+
"commit": { "type": "boolean" },
|
|
171
|
+
"verify": { "type": "boolean" },
|
|
172
|
+
"audit": { "type": "boolean" }
|
|
173
|
+
}
|
|
174
|
+
},
|
|
175
|
+
"idempotency": {
|
|
176
|
+
"type": "object",
|
|
177
|
+
"additionalProperties": false,
|
|
178
|
+
"required": ["required"],
|
|
179
|
+
"properties": {
|
|
180
|
+
"required": { "type": "boolean" },
|
|
181
|
+
"key_fields": {
|
|
182
|
+
"type": "array",
|
|
183
|
+
"items": { "type": "string", "minLength": 1 }
|
|
184
|
+
}
|
|
185
|
+
}
|
|
186
|
+
},
|
|
187
|
+
"observability": {
|
|
188
|
+
"type": "object",
|
|
189
|
+
"additionalProperties": false,
|
|
190
|
+
"required": ["log_inputs", "log_outputs"],
|
|
191
|
+
"properties": {
|
|
192
|
+
"log_inputs": {
|
|
193
|
+
"enum": ["none", "summary", "redacted"]
|
|
194
|
+
},
|
|
195
|
+
"log_outputs": {
|
|
196
|
+
"enum": ["none", "summary", "redacted"]
|
|
197
|
+
},
|
|
198
|
+
"trace_attributes": {
|
|
199
|
+
"type": "object",
|
|
200
|
+
"additionalProperties": {
|
|
201
|
+
"type": ["string", "number", "boolean", "null"]
|
|
202
|
+
}
|
|
203
|
+
}
|
|
204
|
+
}
|
|
205
|
+
},
|
|
206
|
+
"evals": {
|
|
207
|
+
"type": "object",
|
|
208
|
+
"additionalProperties": false,
|
|
209
|
+
"properties": {
|
|
210
|
+
"golden": {
|
|
211
|
+
"type": "array",
|
|
212
|
+
"items": { "type": "string", "minLength": 1 }
|
|
213
|
+
},
|
|
214
|
+
"red_team": {
|
|
215
|
+
"type": "array",
|
|
216
|
+
"items": { "type": "string", "minLength": 1 }
|
|
217
|
+
}
|
|
218
|
+
}
|
|
219
|
+
},
|
|
220
|
+
"extensions": {
|
|
221
|
+
"type": "object"
|
|
222
|
+
}
|
|
223
|
+
},
|
|
224
|
+
"$defs": {
|
|
225
|
+
"json_schema_object": {
|
|
226
|
+
"type": "object",
|
|
227
|
+
"required": ["type"],
|
|
228
|
+
"additionalProperties": true,
|
|
229
|
+
"properties": {
|
|
230
|
+
"type": {
|
|
231
|
+
"type": ["string", "array"]
|
|
232
|
+
}
|
|
233
|
+
}
|
|
234
|
+
},
|
|
235
|
+
"policy_rule": {
|
|
236
|
+
"type": "object",
|
|
237
|
+
"additionalProperties": false,
|
|
238
|
+
"required": ["rule", "reason"],
|
|
239
|
+
"properties": {
|
|
240
|
+
"rule": { "type": "string", "minLength": 1 },
|
|
241
|
+
"reason": { "type": "string", "minLength": 1 },
|
|
242
|
+
"field": { "type": "string" },
|
|
243
|
+
"operator": {
|
|
244
|
+
"enum": ["eq", "neq", "gt", "gte", "lt", "lte", "in", "not_in", "exists"]
|
|
245
|
+
},
|
|
246
|
+
"value": {}
|
|
247
|
+
}
|
|
248
|
+
}
|
|
249
|
+
}
|
|
250
|
+
}
|
|
@@ -0,0 +1,97 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://raw.githubusercontent.com/xennon101/ai-capability-framework/main/schemas/entity-manifest.schema.json",
|
|
4
|
+
"title": "AICF Entity Manifest",
|
|
5
|
+
"type": "object",
|
|
6
|
+
"additionalProperties": false,
|
|
7
|
+
"required": [
|
|
8
|
+
"schema_version",
|
|
9
|
+
"id",
|
|
10
|
+
"version",
|
|
11
|
+
"name",
|
|
12
|
+
"summary",
|
|
13
|
+
"canonical_id",
|
|
14
|
+
"data_classification",
|
|
15
|
+
"lookup",
|
|
16
|
+
"allowed_actions",
|
|
17
|
+
"model_guidance"
|
|
18
|
+
],
|
|
19
|
+
"properties": {
|
|
20
|
+
"schema_version": {
|
|
21
|
+
"const": "1.0"
|
|
22
|
+
},
|
|
23
|
+
"id": {
|
|
24
|
+
"type": "string",
|
|
25
|
+
"pattern": "^[A-Z][A-Za-z0-9_]*$"
|
|
26
|
+
},
|
|
27
|
+
"version": {
|
|
28
|
+
"type": "string",
|
|
29
|
+
"pattern": "^(0|[1-9]\\d*)\\.(0|[1-9]\\d*)\\.(0|[1-9]\\d*)(?:-[0-9A-Za-z-.]+)?(?:\\+[0-9A-Za-z-.]+)?$"
|
|
30
|
+
},
|
|
31
|
+
"name": {
|
|
32
|
+
"type": "string",
|
|
33
|
+
"minLength": 1
|
|
34
|
+
},
|
|
35
|
+
"summary": {
|
|
36
|
+
"type": "string",
|
|
37
|
+
"minLength": 1
|
|
38
|
+
},
|
|
39
|
+
"canonical_id": {
|
|
40
|
+
"type": "object",
|
|
41
|
+
"additionalProperties": false,
|
|
42
|
+
"required": ["field"],
|
|
43
|
+
"properties": {
|
|
44
|
+
"field": { "type": "string", "minLength": 1 },
|
|
45
|
+
"pattern": { "type": "string" }
|
|
46
|
+
}
|
|
47
|
+
},
|
|
48
|
+
"data_classification": {
|
|
49
|
+
"type": "object",
|
|
50
|
+
"additionalProperties": false,
|
|
51
|
+
"required": ["default"],
|
|
52
|
+
"properties": {
|
|
53
|
+
"default": { "type": "string", "minLength": 1 },
|
|
54
|
+
"fields": {
|
|
55
|
+
"type": "object",
|
|
56
|
+
"additionalProperties": { "type": "string" }
|
|
57
|
+
}
|
|
58
|
+
}
|
|
59
|
+
},
|
|
60
|
+
"relationships": {
|
|
61
|
+
"type": "array",
|
|
62
|
+
"items": {
|
|
63
|
+
"type": "object",
|
|
64
|
+
"additionalProperties": false,
|
|
65
|
+
"required": ["name", "target_entity", "cardinality"],
|
|
66
|
+
"properties": {
|
|
67
|
+
"name": { "type": "string", "minLength": 1 },
|
|
68
|
+
"target_entity": { "type": "string", "minLength": 1 },
|
|
69
|
+
"cardinality": {
|
|
70
|
+
"enum": ["one_to_one", "one_to_many", "many_to_one", "many_to_many"]
|
|
71
|
+
},
|
|
72
|
+
"lookup_capability": { "type": "string" }
|
|
73
|
+
}
|
|
74
|
+
}
|
|
75
|
+
},
|
|
76
|
+
"lookup": {
|
|
77
|
+
"type": "object",
|
|
78
|
+
"additionalProperties": false,
|
|
79
|
+
"required": ["primary_capability"],
|
|
80
|
+
"properties": {
|
|
81
|
+
"primary_capability": { "type": "string" }
|
|
82
|
+
}
|
|
83
|
+
},
|
|
84
|
+
"allowed_actions": {
|
|
85
|
+
"type": "array",
|
|
86
|
+
"minItems": 1,
|
|
87
|
+
"items": { "type": "string", "minLength": 1 }
|
|
88
|
+
},
|
|
89
|
+
"model_guidance": {
|
|
90
|
+
"type": "string",
|
|
91
|
+
"minLength": 1
|
|
92
|
+
},
|
|
93
|
+
"extensions": {
|
|
94
|
+
"type": "object"
|
|
95
|
+
}
|
|
96
|
+
}
|
|
97
|
+
}
|
|
@@ -0,0 +1,134 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://raw.githubusercontent.com/xennon101/ai-capability-framework/main/schemas/eval-case.schema.json",
|
|
4
|
+
"title": "AICF Eval Case",
|
|
5
|
+
"type": "object",
|
|
6
|
+
"additionalProperties": false,
|
|
7
|
+
"required": ["schema_version", "id", "input", "expected", "scorers"],
|
|
8
|
+
"properties": {
|
|
9
|
+
"schema_version": {
|
|
10
|
+
"const": "1.0"
|
|
11
|
+
},
|
|
12
|
+
"id": {
|
|
13
|
+
"type": "string",
|
|
14
|
+
"pattern": "^[a-z][a-z0-9_]*(\\.[a-z][a-z0-9_]*)*$"
|
|
15
|
+
},
|
|
16
|
+
"name": {
|
|
17
|
+
"type": "string"
|
|
18
|
+
},
|
|
19
|
+
"capability_under_test": {
|
|
20
|
+
"type": "string"
|
|
21
|
+
},
|
|
22
|
+
"tags": {
|
|
23
|
+
"type": "array",
|
|
24
|
+
"items": { "type": "string" }
|
|
25
|
+
},
|
|
26
|
+
"input": {
|
|
27
|
+
"type": "object",
|
|
28
|
+
"additionalProperties": false,
|
|
29
|
+
"required": ["user_message"],
|
|
30
|
+
"properties": {
|
|
31
|
+
"user_message": { "type": "string", "minLength": 1 },
|
|
32
|
+
"conversation": {
|
|
33
|
+
"type": "array",
|
|
34
|
+
"items": {
|
|
35
|
+
"type": "object",
|
|
36
|
+
"additionalProperties": false,
|
|
37
|
+
"required": ["role", "content"],
|
|
38
|
+
"properties": {
|
|
39
|
+
"role": { "enum": ["user", "assistant", "system", "tool"] },
|
|
40
|
+
"content": { "type": "string" }
|
|
41
|
+
}
|
|
42
|
+
}
|
|
43
|
+
}
|
|
44
|
+
}
|
|
45
|
+
},
|
|
46
|
+
"context": {
|
|
47
|
+
"type": "object"
|
|
48
|
+
},
|
|
49
|
+
"expected": { "$ref": "#/$defs/expected_behavior" },
|
|
50
|
+
"scorers": {
|
|
51
|
+
"type": "array",
|
|
52
|
+
"minItems": 1,
|
|
53
|
+
"items": {
|
|
54
|
+
"type": "object",
|
|
55
|
+
"additionalProperties": true,
|
|
56
|
+
"required": ["type"],
|
|
57
|
+
"properties": {
|
|
58
|
+
"type": { "type": "string", "minLength": 1 }
|
|
59
|
+
}
|
|
60
|
+
}
|
|
61
|
+
},
|
|
62
|
+
"extensions": {
|
|
63
|
+
"type": "object"
|
|
64
|
+
}
|
|
65
|
+
},
|
|
66
|
+
"$defs": {
|
|
67
|
+
"expected_behavior": {
|
|
68
|
+
"type": "object",
|
|
69
|
+
"additionalProperties": false,
|
|
70
|
+
"minProperties": 1,
|
|
71
|
+
"properties": {
|
|
72
|
+
"selected_capabilities": {
|
|
73
|
+
"type": "object",
|
|
74
|
+
"additionalProperties": false,
|
|
75
|
+
"properties": {
|
|
76
|
+
"includes": {
|
|
77
|
+
"type": "array",
|
|
78
|
+
"items": { "type": "string", "minLength": 1 }
|
|
79
|
+
},
|
|
80
|
+
"excludes": {
|
|
81
|
+
"type": "array",
|
|
82
|
+
"items": { "type": "string", "minLength": 1 }
|
|
83
|
+
}
|
|
84
|
+
}
|
|
85
|
+
},
|
|
86
|
+
"tool_calls": {
|
|
87
|
+
"type": "array",
|
|
88
|
+
"items": {
|
|
89
|
+
"type": "object",
|
|
90
|
+
"additionalProperties": false,
|
|
91
|
+
"required": ["capability_id"],
|
|
92
|
+
"properties": {
|
|
93
|
+
"capability_id": { "type": "string", "minLength": 1 },
|
|
94
|
+
"args_match": { "type": "object" }
|
|
95
|
+
}
|
|
96
|
+
}
|
|
97
|
+
},
|
|
98
|
+
"policy_decision": {
|
|
99
|
+
"enum": ["allowed", "approval_required", "denied"]
|
|
100
|
+
},
|
|
101
|
+
"action_state": {
|
|
102
|
+
"enum": ["none", "prepared", "approval_required", "committed", "denied", "refused"]
|
|
103
|
+
},
|
|
104
|
+
"no_commit": { "type": "boolean" },
|
|
105
|
+
"refusal": {
|
|
106
|
+
"type": "object",
|
|
107
|
+
"additionalProperties": false,
|
|
108
|
+
"required": ["required"],
|
|
109
|
+
"properties": {
|
|
110
|
+
"required": { "type": "boolean" },
|
|
111
|
+
"reason_contains": {
|
|
112
|
+
"type": "array",
|
|
113
|
+
"items": { "type": "string", "minLength": 1 }
|
|
114
|
+
}
|
|
115
|
+
}
|
|
116
|
+
},
|
|
117
|
+
"response": {
|
|
118
|
+
"type": "object",
|
|
119
|
+
"additionalProperties": false,
|
|
120
|
+
"properties": {
|
|
121
|
+
"must_include": {
|
|
122
|
+
"type": "array",
|
|
123
|
+
"items": { "type": "string", "minLength": 1 }
|
|
124
|
+
},
|
|
125
|
+
"must_not_include": {
|
|
126
|
+
"type": "array",
|
|
127
|
+
"items": { "type": "string", "minLength": 1 }
|
|
128
|
+
}
|
|
129
|
+
}
|
|
130
|
+
}
|
|
131
|
+
}
|
|
132
|
+
}
|
|
133
|
+
}
|
|
134
|
+
}
|
|
@@ -0,0 +1,77 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://raw.githubusercontent.com/xennon101/ai-capability-framework/main/schemas/eval-result.schema.json",
|
|
4
|
+
"title": "AICF Eval Result Fixture",
|
|
5
|
+
"type": "object",
|
|
6
|
+
"additionalProperties": false,
|
|
7
|
+
"required": ["schema_version", "results"],
|
|
8
|
+
"properties": {
|
|
9
|
+
"schema_version": {
|
|
10
|
+
"const": "1.0"
|
|
11
|
+
},
|
|
12
|
+
"results": {
|
|
13
|
+
"type": "array",
|
|
14
|
+
"items": { "$ref": "#/$defs/eval_candidate_result" }
|
|
15
|
+
},
|
|
16
|
+
"extensions": {
|
|
17
|
+
"type": "object"
|
|
18
|
+
}
|
|
19
|
+
},
|
|
20
|
+
"$defs": {
|
|
21
|
+
"eval_candidate_result": {
|
|
22
|
+
"type": "object",
|
|
23
|
+
"additionalProperties": false,
|
|
24
|
+
"required": ["eval_id"],
|
|
25
|
+
"properties": {
|
|
26
|
+
"eval_id": {
|
|
27
|
+
"type": "string",
|
|
28
|
+
"minLength": 1
|
|
29
|
+
},
|
|
30
|
+
"selected_capabilities": {
|
|
31
|
+
"type": "array",
|
|
32
|
+
"items": { "type": "string", "minLength": 1 }
|
|
33
|
+
},
|
|
34
|
+
"tool_calls": {
|
|
35
|
+
"type": "array",
|
|
36
|
+
"items": {
|
|
37
|
+
"type": "object",
|
|
38
|
+
"additionalProperties": false,
|
|
39
|
+
"required": ["capability_id"],
|
|
40
|
+
"properties": {
|
|
41
|
+
"capability_id": { "type": "string", "minLength": 1 },
|
|
42
|
+
"args": { "type": "object" }
|
|
43
|
+
}
|
|
44
|
+
}
|
|
45
|
+
},
|
|
46
|
+
"policy_decision": {
|
|
47
|
+
"enum": ["allowed", "approval_required", "denied"]
|
|
48
|
+
},
|
|
49
|
+
"action_state": {
|
|
50
|
+
"enum": ["none", "prepared", "approval_required", "committed", "denied", "refused"]
|
|
51
|
+
},
|
|
52
|
+
"committed_capabilities": {
|
|
53
|
+
"type": "array",
|
|
54
|
+
"items": { "type": "string", "minLength": 1 }
|
|
55
|
+
},
|
|
56
|
+
"refusal": {
|
|
57
|
+
"type": "object",
|
|
58
|
+
"additionalProperties": false,
|
|
59
|
+
"properties": {
|
|
60
|
+
"present": { "type": "boolean" },
|
|
61
|
+
"reason": { "type": "string" }
|
|
62
|
+
}
|
|
63
|
+
},
|
|
64
|
+
"response": {
|
|
65
|
+
"type": "object",
|
|
66
|
+
"additionalProperties": false,
|
|
67
|
+
"properties": {
|
|
68
|
+
"text": { "type": "string" }
|
|
69
|
+
}
|
|
70
|
+
},
|
|
71
|
+
"extensions": {
|
|
72
|
+
"type": "object"
|
|
73
|
+
}
|
|
74
|
+
}
|
|
75
|
+
}
|
|
76
|
+
}
|
|
77
|
+
}
|