rlhf-feedback-loop 0.5.0

package/adapters/chatgpt/openapi.yaml

@@ -0,0 +1,292 @@
+openapi: 3.1.0
+info:
+  title: RLHF Feedback Loop API
+  version: 1.1.0
+  description: |
+    Production API for feedback capture, schema-validated memory promotion,
+    prevention rule generation, and DPO export.
+servers:
+  - url: http://localhost:8787
+security:
+  - bearerAuth: []
+components:
+  securitySchemes:
+    bearerAuth:
+      type: http
+      scheme: bearer
+      bearerFormat: API Key
+  schemas:
+    RubricScore:
+      type: object
+      required: [criterion, score]
+      properties:
+        criterion:
+          type: string
+        score:
+          type: number
+          minimum: 1
+          maximum: 5
+        evidence:
+          type: string
+        judge:
+          type: string
+    CaptureFeedbackRequest:
+      type: object
+      required: [signal, context]
+      properties:
+        signal:
+          type: string
+          enum: [up, down, positive, negative]
+        context:
+          type: string
+        whatWentWrong:
+          type: string
+        whatToChange:
+          type: string
+        whatWorked:
+          type: string
+        rubricScores:
+          type: array
+          items:
+            $ref: '#/components/schemas/RubricScore'
+        guardrails:
+          type: object
+          properties:
+            testsPassed:
+              type: boolean
+            pathSafety:
+              type: boolean
+            budgetCompliant:
+              type: boolean
+        tags:
+          oneOf:
+            - type: array
+              items:
+                type: string
+            - type: string
+        skill:
+          type: string
+    IntentPlanRequest:
+      type: object
+      required: [intentId]
+      properties:
+        intentId:
+          type: string
+        context:
+          type: string
+        mcpProfile:
+          type: string
+        bundleId:
+          type: string
+        approved:
+          type: boolean
+paths:
+  /healthz:
+    get:
+      operationId: healthz
+      responses:
+        '200':
+          description: Service health
+        '401':
+          description: Unauthorized
+  /v1/feedback/capture:
+    post:
+      operationId: captureFeedback
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/CaptureFeedbackRequest'
+      responses:
+        '200':
+          description: Feedback accepted and promoted to memory
+        '422':
+          description: Feedback recorded but rejected for memory promotion
+        '401':
+          description: Unauthorized
+  /v1/feedback/stats:
+    get:
+      operationId: getFeedbackStats
+      responses:
+        '200':
+          description: Aggregated feedback statistics
+        '401':
+          description: Unauthorized
+  /v1/intents/catalog:
+    get:
+      operationId: listIntentCatalog
+      parameters:
+        - in: query
+          name: mcpProfile
+          schema:
+            type: string
+        - in: query
+          name: bundleId
+          schema:
+            type: string
+      responses:
+        '200':
+          description: Intent catalog with risk and checkpoint metadata
+        '401':
+          description: Unauthorized
+  /v1/intents/plan:
+    post:
+      operationId: planIntent
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/IntentPlanRequest'
+      responses:
+        '200':
+          description: Policy-scoped intent execution plan
+        '400':
+          description: Invalid intent request
+        '401':
+          description: Unauthorized
+  /v1/feedback/summary:
+    get:
+      operationId: getFeedbackSummary
+      parameters:
+        - in: query
+          name: recent
+          schema:
+            type: integer
+            default: 20
+      responses:
+        '200':
+          description: Feedback summary text
+        '401':
+          description: Unauthorized
+  /v1/feedback/rules:
+    post:
+      operationId: generatePreventionRules
+      requestBody:
+        required: false
+        content:
+          application/json:
+            schema:
+              type: object
+              properties:
+                minOccurrences:
+                  type: integer
+                  default: 2
+                outputPath:
+                  type: string
+      responses:
+        '200':
+          description: Prevention rules generated
+        '401':
+          description: Unauthorized
+  /v1/dpo/export:
+    post:
+      operationId: exportDpoPairs
+      requestBody:
+        required: false
+        content:
+          application/json:
+            schema:
+              type: object
+              properties:
+                inputPath:
+                  type: string
+                memoryLogPath:
+                  type: string
+                outputPath:
+                  type: string
+      responses:
+        '200':
+          description: DPO export completed
+        '401':
+          description: Unauthorized
+  /v1/context/construct:
+    post:
+      operationId: constructContextPack
+      requestBody:
+        required: false
+        content:
+          application/json:
+            schema:
+              type: object
+              properties:
+                query:
+                  type: string
+                maxItems:
+                  type: integer
+                  default: 8
+                maxChars:
+                  type: integer
+                  default: 6000
+                namespaces:
+                  type: array
+                  items:
+                    type: string
+                    enum:
+                      - raw_history
+                      - memory/error
+                      - memory/learning
+                      - rules
+                      - tools
+                      - provenance
+      responses:
+        '200':
+          description: Context pack created
+        '400':
+          description: Invalid namespace selection
+        '401':
+          description: Unauthorized
+  /v1/context/evaluate:
+    post:
+      operationId: evaluateContextPack
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              type: object
+              required: [packId, outcome]
+              properties:
+                packId:
+                  type: string
+                outcome:
+                  type: string
+                signal:
+                  type: string
+                notes:
+                  type: string
+                rubricScores:
+                  type: array
+                  items:
+                    $ref: '#/components/schemas/RubricScore'
+                guardrails:
+                  type: object
+                  properties:
+                    testsPassed:
+                      type: boolean
+                    pathSafety:
+                      type: boolean
+                    budgetCompliant:
+                      type: boolean
+      responses:
+        '200':
+          description: Evaluation recorded
+        '400':
+          description: Invalid rubric payload
+        '401':
+          description: Unauthorized
+  /v1/context/provenance:
+    get:
+      operationId: getContextProvenance
+      parameters:
+        - in: query
+          name: limit
+          schema:
+            type: integer
+            default: 50
+      responses:
+        '200':
+          description: Recent provenance events
+        '401':
+          description: Unauthorized

package/adapters/claude/.mcp.json

@@ -0,0 +1,8 @@
+{
+  "mcpServers": {
+    "rlhf-feedback-loop": {
+      "command": "node",
+      "args": ["adapters/mcp/server-stdio.js"]
+    }
+  }
+}

package/adapters/codex/config.toml

@@ -0,0 +1,4 @@
+# Codex MCP profile (copy into ~/.codex/config.toml or merge section)
+[mcp_servers.rlhf_feedback_loop]
+command = "node"
+args = ["adapters/mcp/server-stdio.js"]

package/adapters/gemini/function-declarations.json

@@ -0,0 +1,95 @@
+{
+  "tools": [
+    {
+      "name": "capture_feedback",
+      "description": "Capture thumbs-up/down feedback and promote actionable memories",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "signal": { "type": "string", "enum": ["up", "down"] },
+          "context": { "type": "string" },
+          "whatWentWrong": { "type": "string" },
+          "whatToChange": { "type": "string" },
+          "whatWorked": { "type": "string" },
+          "rubricScores": {
+            "type": "array",
+            "items": {
+              "type": "object",
+              "properties": {
+                "criterion": { "type": "string" },
+                "score": { "type": "number" },
+                "evidence": { "type": "string" },
+                "judge": { "type": "string" }
+              }
+            }
+          },
+          "guardrails": {
+            "type": "object",
+            "properties": {
+              "testsPassed": { "type": "boolean" },
+              "pathSafety": { "type": "boolean" },
+              "budgetCompliant": { "type": "boolean" }
+            }
+          },
+          "tags": {
+            "type": "array",
+            "items": { "type": "string" }
+          },
+          "skill": { "type": "string" }
+        },
+        "required": ["signal", "context"]
+      },
+      "http": {
+        "method": "POST",
+        "path": "/v1/feedback/capture"
+      }
+    },
+    {
+      "name": "feedback_summary",
+      "description": "Get a compact summary of recent feedback performance",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "recent": { "type": "integer" }
+        }
+      },
+      "http": {
+        "method": "GET",
+        "path": "/v1/feedback/summary"
+      }
+    },
+    {
+      "name": "prevention_rules",
+      "description": "Generate prevention rules from repeated mistakes",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "minOccurrences": { "type": "integer" }
+        }
+      },
+      "http": {
+        "method": "POST",
+        "path": "/v1/feedback/rules"
+      }
+    },
+    {
+      "name": "plan_intent",
+      "description": "Generate a policy-aware execution plan with checkpoint requirements",
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "intentId": { "type": "string" },
+          "context": { "type": "string" },
+          "mcpProfile": { "type": "string" },
+          "bundleId": { "type": "string" },
+          "approved": { "type": "boolean" }
+        },
+        "required": ["intentId"]
+      },
+      "http": {
+        "method": "POST",
+        "path": "/v1/intents/plan"
+      }
+    }
+  ]
+}