@framers/agentos-skills-registry 0.7.0 → 0.8.0

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@framers/agentos-skills-registry",
3
- "version": "0.7.0",
3
+ "version": "0.8.0",
4
4
  "files": [
5
5
  "dist",
6
6
  "registry",
@@ -0,0 +1,38 @@
1
+ ---
2
+ name: grounding-guard
3
+ version: '1.0.0'
4
+ description: Verify response faithfulness against RAG source documents using NLI entailment and LLM-as-judge
5
+ author: Frame.dev
6
+ namespace: wunderland
7
+ category: security
8
+ tags: [guardrails, hallucination, grounding, faithfulness, nli, rag, fact-checking]
9
+ requires_tools: [check_grounding]
10
+ metadata:
11
+ agentos:
12
+ emoji: "\U0001F50D"
13
+ ---
14
+
15
+ # Grounding Guard
16
+
17
+ A guardrail automatically verifies that your responses are faithful to
18
+ the source documents retrieved via RAG. Claims in your output are checked
19
+ against the retrieved sources using NLI entailment detection.
20
+
21
+ ## When to Use check_grounding
22
+
23
+ - Before presenting synthesized answers from multiple RAG sources
24
+ - To verify that summarized content faithfully represents the originals
25
+ - When combining information from multiple documents
26
+
27
+ ## What It Checks
28
+
29
+ - **Supported**: claim is entailed by at least one source document
30
+ - **Contradicted**: claim directly contradicts a source document
31
+ - **Unverifiable**: claim cannot be found in any source (potential hallucination)
32
+
33
+ ## Constraints
34
+
35
+ - Only runs when RAG sources are available (no sources → no verification)
36
+ - NLI model (~40MB) loads lazily on first grounding check
37
+ - LLM escalation for ambiguous claims adds ~150-500ms (only when configured)
38
+ - Best for factual/informational responses; less useful for creative/opinion content