feat: conformance scenarios (core-01..core-09)

Declarative conformance scenarios for aicc/0.1 covering: handshake, manifest_request, tool call ok/unknown/execution-failed, protocol mismatch, heartbeat, session close, event delivery. Plus runner documentation (conformance/README.md).
2026-08-08 03:27:20 +03:00
parent e75ba873d8
commit fa5ef35bff
10 changed files with 305 additions and 0 deletions
@@ -0,0 +1,74 @@
+# AICC Conformance Scenarios
+
+This directory contains declarative conformance scenarios for the AICC Protocol (aicc/0.1).
+
+Each JSON file describes a sequence of messages exchanged with a bridge and
+the expected responses. Any bridge that passes all scenarios in this
+directory satisfies the conformance criteria in §13 of `../core.md`.
+
+## Scenario format
+
+```json
+{
+  "id": "core-01-handshake",
+  "title": "Human readable title",
+  "category": "core",
+  "version": "aicc/0.1",
+  "steps": [
+    { "send": { "type": "tool_call", "tool": "echo", "input": {} } },
+    { "expect": { "type": "tool_result", "ok": true, "call_id_echo": true } }
+  ]
+}
+```
+
+### Steps
+
+| Step kind | Semantics |
+|-----------|-----------|
+| `send`    | Send a message to the bridge. `session_id` and `message_id` are filled in by the runner if absent. |
+| `expect`  | Wait for the next message and assert it matches the expectation. |
+
+### Expectation matching
+
+Expectations are matched **subset-wise**: every field present in the
+expectation must match the received message; extra fields in the received
+message are allowed.
+
+Special matchers:
+
+| Field | Meaning |
+|-------|---------|
+| `call_id_echo: true` | Received `call_id` must equal the `call_id` of the last `send`. |
+| `session_id_echo: true` | Received `session_id` must equal the session's id. |
+| `{"$kind": "uuid"}` | Field must look like a UUID. |
+| `{"$kind": "integer"}` / `{"$kind": "array"}` | Field must be of that JSON type. |
+| `{"$gt": N}` | Field must be numerically greater than N. |
+| `{"$enum": [...]}` | Field must be one of the listed values. |
+| `{"$required": [...]}` | Object field must contain these keys. |
+
+## Running
+
+Reference runner: `aicc-py` (see https://github.com/emil28092005/aicc-py).
+Run `python -m aicc.conformance <scenarios-dir>` or the pytest suite.
+
+## Scenario inventory
+
+| ID | Checks |
+|----|--------|
+| core-01-handshake | Valid session_init on connect |
+| core-02-manifest-request | manifest_request returns manifest |
+| core-03-tool-call-ok | Valid tool call succeeds, call_id echoed |
+| core-04-tool-unknown | Unknown tool → tool_unknown |
+| core-05-tool-execution-failed | Throwing tool → execution_failed |
+| core-06-protocol-mismatch | Wrong protocol → protocol_mismatch |
+| core-07-heartbeat | Heartbeat acknowledged |
+| core-08-session-close | Close ends session, then session_expired |
+| core-09-event-delivery | Environment events delivered |
+
+## Adding scenarios
+
+1. Create `N-<name>.json` following the format above.
+2. Add a row to the inventory table.
+3. Run the runner against the reference bridge in aicc-py.
+4. Scenarios requiring tools not in the reference bridge must document
+   the requirement in `description`.
@@ -0,0 +1,21 @@
+{
+  "id": "core-01-handshake",
+  "title": "Handshake: bridge sends a valid session_init",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "On connect, the bridge must push a session_init message with a valid envelope and manifest structure.",
+  "steps": [
+    {
+      "expect": {
+        "type": "session_init",
+        "protocol": "aicc/0.1",
+        "session_id": {"$kind": "uuid"},
+        "tick_rate_hz": {"$gt": 0},
+        "tick_mode": {"$enum": ["fixed", "event", "hybrid"]},
+        "world": {"$required": ["name", "kind"]},
+        "capabilities": {"$required": ["sensors", "actuators", "generators"]},
+        "tools": {"$kind": "array"}
+      }
+    }
+  ]
+}
@@ -0,0 +1,20 @@
+{
+  "id": "core-02-manifest-request",
+  "title": "manifest_request returns the same session_init",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "A manifest_request must be answered with the session manifest for the current session.",
+  "steps": [
+    {
+      "send": {
+        "type": "manifest_request"
+      }
+    },
+    {
+      "expect": {
+        "type": "session_init",
+        "session_id_echo": true
+      }
+    }
+  ]
+}
@@ -0,0 +1,24 @@
+{
+  "id": "core-03-tool-call-ok",
+  "title": "Valid tool call returns ok:true with echo call_id",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "A tool registered in the manifest must execute and return ok:true, echoing the caller's call_id.",
+  "steps": [
+    {
+      "send": {
+        "type": "tool_call",
+        "tool": "echo",
+        "input": {"value": "hello"}
+      }
+    },
+    {
+      "expect": {
+        "type": "tool_result",
+        "ok": true,
+        "call_id_echo": true,
+        "output": {"value": "hello"}
+      }
+    }
+  ]
+}
@@ -0,0 +1,27 @@
+{
+  "id": "core-04-tool-unknown",
+  "title": "Unknown tool returns tool_unknown",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "Calling a tool that is not in the manifest must fail with error code tool_unknown and retryable:false.",
+  "steps": [
+    {
+      "send": {
+        "type": "tool_call",
+        "tool": "definitely_not_a_tool",
+        "input": {}
+      }
+    },
+    {
+      "expect": {
+        "type": "tool_result",
+        "ok": false,
+        "call_id_echo": true,
+        "error": {
+          "code": "tool_unknown",
+          "retryable": false
+        }
+      }
+    }
+  ]
+}
@@ -0,0 +1,27 @@
+{
+  "id": "core-05-tool-execution-failed",
+  "title": "Tool that raises returns execution_failed",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "A tool implementation that throws must be caught by the bridge and surfaced as error code execution_failed, not crash the session.",
+  "steps": [
+    {
+      "send": {
+        "type": "tool_call",
+        "tool": "boom",
+        "input": {}
+      }
+    },
+    {
+      "expect": {
+        "type": "tool_result",
+        "ok": false,
+        "call_id_echo": true,
+        "error": {
+          "code": "execution_failed",
+          "retryable": false
+        }
+      }
+    }
+  ]
+}
@@ -0,0 +1,26 @@
+{
+  "id": "core-06-protocol-mismatch",
+  "title": "Protocol mismatch is rejected",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "A message with a wrong protocol version must be rejected with protocol_mismatch.",
+  "steps": [
+    {
+      "send": {
+        "protocol": "aicc/0.2",
+        "type": "tool_call",
+        "tool": "echo",
+        "input": {}
+      }
+    },
+    {
+      "expect": {
+        "type": "error",
+        "error": {
+          "code": "protocol_mismatch",
+          "retryable": false
+        }
+      }
+    }
+  ]
+}
@@ -0,0 +1,19 @@
+{
+  "id": "core-07-heartbeat",
+  "title": "Heartbeat is answered with heartbeat",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "A heartbeat must be acknowledged with a heartbeat, keeping the session alive.",
+  "steps": [
+    {
+      "send": {
+        "type": "heartbeat"
+      }
+    },
+    {
+      "expect": {
+        "type": "heartbeat"
+      }
+    }
+  ]
+}
@@ -0,0 +1,36 @@
+{
+  "id": "core-08-session-close",
+  "title": "session_close ends the session",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "After session_close, further tool calls must fail with session_expired.",
+  "steps": [
+    {
+      "send": {
+        "type": "session_close",
+        "reason": "conformance"
+      }
+    },
+    {
+      "expect": {
+        "type": "session_close"
+      }
+    },
+    {
+      "send": {
+        "type": "tool_call",
+        "tool": "echo",
+        "input": {}
+      }
+    },
+    {
+      "expect": {
+        "type": "error",
+        "error": {
+          "code": "session_expired",
+          "retryable": false
+        }
+      }
+    }
+  ]
+}
@@ -0,0 +1,31 @@
+{
+  "id": "core-09-event-delivery",
+  "title": "Environment events are delivered to the agent",
+  "category": "core",
+  "version": "aicc/0.1",
+  "description": "Events pushed by the bridge (e.g. collision) arrive as event messages with a tick in meta.",
+  "steps": [
+    {
+      "send": {
+        "type": "tool_call",
+        "tool": "bump",
+        "input": {}
+      }
+    },
+    {
+      "expect": {
+        "type": "tool_result",
+        "ok": true
+      }
+    },
+    {
+      "expect": {
+        "type": "event",
+        "topic": "collision",
+        "meta": {
+          "tick": {"$kind": "integer"}
+        }
+      }
+    }
+  ]
+}