mcp-nervous-system 1.2.0 → 1.3.0
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/README.md +20 -5
- package/package.json +2 -2
- package/server.js +918 -477
package/README.md
CHANGED
|
@@ -4,7 +4,7 @@
|
|
|
4
4
|
|
|
5
5
|
7 mechanically enforced rules that prevent the most common failure modes when LLMs have access to real infrastructure: context loss, silent failures, file damage, goal drift, and overreach.
|
|
6
6
|
|
|
7
|
-
Built by [Arthur Palyan](https://www.levelsofself.com) at Palyan AI.
|
|
7
|
+
Built by [Arthur Palyan](https://www.levelsofself.com) at Palyan AI. 14 tools including task complexity classification and user intent parsing. Battle-tested on an 11-member AI family running 28 processes 24/7 on a single VPS. 58+ violations logged, 0 bypassed.
|
|
8
8
|
|
|
9
9
|
## The Problem
|
|
10
10
|
|
|
@@ -60,6 +60,18 @@ Protocol: MCP 2024-11-05 (Streamable HTTP + SSE)
|
|
|
60
60
|
Authentication: None required
|
|
61
61
|
```
|
|
62
62
|
|
|
63
|
+
## NEW in v1.3.0
|
|
64
|
+
|
|
65
|
+
**classify_task_complexity** (free tier)
|
|
66
|
+
Analyzes any task across 6 dimensions (scope, judgment, risk, context depth, ambiguity, verification difficulty) and recommends the optimal model tier. Routes simple tasks to Haiku (~90% savings), standard tasks to Sonnet (~60% savings), and complex reasoning to Opus. Pass optional structured hints for higher accuracy.
|
|
67
|
+
|
|
68
|
+
**parse_user_intent** (free tier)
|
|
69
|
+
Decomposes user requests into numbered deliverables with confidence scoring. Below 80% confidence: the model should clarify before executing, not guess. Detects ambiguity, unresolved pronouns, hedging language, and implicit expectations. The 80% rule: understand first, execute second.
|
|
70
|
+
|
|
71
|
+
These tools work together. parse_user_intent breaks down what the person wants. classify_task_complexity routes each piece to the right model. The Nervous System enforces the rules while execution happens.
|
|
72
|
+
|
|
73
|
+
**Positioning: Auto mode (launching March 12) decides what Claude CAN do. The Nervous System governs HOW it behaves while doing it.**
|
|
74
|
+
|
|
63
75
|
## The 7 Rules
|
|
64
76
|
|
|
65
77
|
| # | Rule | What It Prevents |
|
|
@@ -72,7 +84,7 @@ Authentication: None required
|
|
|
72
84
|
| 6 | **Ask Before Touching** | Unauthorized changes. Logic changes need human approval. |
|
|
73
85
|
| 7 | **Hand Off** | Context loss. Written handoffs every 3-4 exchanges. |
|
|
74
86
|
|
|
75
|
-
## MCP Tools (
|
|
87
|
+
## MCP Tools (14)
|
|
76
88
|
|
|
77
89
|
| Tool | Description |
|
|
78
90
|
|------|------------|
|
|
@@ -87,6 +99,9 @@ Authentication: None required
|
|
|
87
99
|
| `emergency_kill_switch` | Emergency shutdown of all PM2 processes. Requires kill switch secret. Logs to tamper-evident audit trail |
|
|
88
100
|
| `verify_audit_chain` | Walks the SHA-256 hash-chained audit log and verifies every entry. Returns chain integrity status |
|
|
89
101
|
| `dispatch_to_llm` | Spawns a background LLM agent to handle a task. Checks RAM, enforces max 2 concurrent dispatches |
|
|
102
|
+
| `classify_task_complexity` | Analyzes task across 6 dimensions, recommends optimal model tier (Haiku/Sonnet/Opus) |
|
|
103
|
+
| `parse_user_intent` | Decomposes requests into numbered deliverables with confidence scoring. 80% gate |
|
|
104
|
+
| `get_positioning` | Returns competitive positioning and differentiation messaging |
|
|
90
105
|
|
|
91
106
|
## Kill Switch
|
|
92
107
|
|
|
@@ -102,7 +117,7 @@ Every guardrail violation, kill switch activation, and dispatch event is recorde
|
|
|
102
117
|
|
|
103
118
|
- Use `verify_audit_chain` to walk the entire chain and verify integrity
|
|
104
119
|
- Returns: valid/invalid status, entry count, and break point if tampered
|
|
105
|
-
-
|
|
120
|
+
- 58+ violations logged, 0 bypassed, 0 chain breaks
|
|
106
121
|
|
|
107
122
|
## Dispatch to LLM
|
|
108
123
|
|
|
@@ -130,11 +145,11 @@ https://api.100levelup.com/family/eu-ai-act.html
|
|
|
130
145
|
|
|
131
146
|
From the live Palyan AI deployment (Feb 28 - Mar 5, 2026):
|
|
132
147
|
|
|
133
|
-
- **
|
|
148
|
+
- **58+** violations caught
|
|
134
149
|
- **29** edits blocked by preflight
|
|
135
150
|
- **13** unique files protected
|
|
136
151
|
- **0** rules bypassed
|
|
137
|
-
- **
|
|
152
|
+
- **28** processes monitored
|
|
138
153
|
- **7** days continuous operation
|
|
139
154
|
|
|
140
155
|
## Live Demo
|
package/package.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "mcp-nervous-system",
|
|
3
|
-
"version": "1.
|
|
4
|
-
"description": "The Nervous System - LLM Behavioral Enforcement Framework. 7 mechanically enforced rules,
|
|
3
|
+
"version": "1.3.0",
|
|
4
|
+
"description": "The Nervous System - LLM Behavioral Enforcement Framework. 7 mechanically enforced rules, 14 tools including task complexity classification and user intent parsing. MCP server for Claude Desktop, Claude Code, and any MCP-compatible client.",
|
|
5
5
|
"main": "server.js",
|
|
6
6
|
"bin": {
|
|
7
7
|
"mcp-nervous-system": "./stdio.js"
|