Releases: cluster-127/tripwired
Releases · cluster-127/tripwired
FilterConfig + QA patterns
Full Changelog: v0.1.6...v0.1.7
Tiered Pattern System
Full Changelog: v0.1.5...v0.1.6
Dangerous command patterns
Full Changelog: v0.1.4...v0.1.5
DriftMonitor implementation + Core module tests
Full Changelog: v0.1.3...v0.1.4
Critical QA fixes - QA report fixes - DriftMonitor, Named Pipe, RegexSet
Critical QA fixes - Real SHA-256 + FAIL on parse error
Rust Execution Kernel
Tripwired v0.1.1 Release Notes
Rust Execution Kernel
Tripwired now achieves 164ms latency for real-time anomaly detection!
New Features
| Feature | Description |
|---|---|
| Rust Sidecar | Zero-GC, deterministic latency kernel |
| Dual IPC | Named Pipe (Windows) / Unix Socket (Linux) + TCP fallback |
| Regex Pre-Filter | 3μs instant bypass for safe logs |
| Audit Trail | Immutable JSONL with model fingerprinting |
| Node.js SDK | Auto-reconnecting client with fallback support |
Performance Benchmark
Before (Node.js spawn): 540ms
After (Rust kernel): 164ms
Improvement: 70% faster 🔥| Scenario | Latency |
|---|---|
| Pre-filtered (safe) | 0.003ms |
| Cold start | 467ms |
| Warm (anomaly) | 164ms |
Architecture
Node.js Bot
↓
[Named Pipe / TCP]
↓
Rust Kernel
→ Regex (3μs) → Safe? → SUSTAIN
→ LLM (164ms) → Anomaly? → KILL → SIGKILL
→ Audit Trail (JSONL)
Technical Details
- TCP Optimizations: tcp_nodelay, no_proxy, pool_idle_timeout(None)
- Model Fingerprinting: SHA-256 config hash for audit compliance
- Prompt Versioning: Hash-based prompt tracking for replay parity
Files Added
kernel/- Complete Rust sidecar binaryCHANGELOG.md- Release history- Updated
README.mdwith architecture and benchmarks
Full Changelog: v0.1.0...v0.1.1
v0.1.0 - Behavioral Control Kernel for AI Agents
Tripwire is a Kill-Switch for autonomous AI agents. It monitors agent behavior and triggers intervention when control degrades.
Features
- ActivityEngine: Detects IDLE, WORKING, LOOPING, RUNAWAY modes
- IntentCore: Produces CONTINUE, PAUSE, STOP decisions
- SafetyGate: Token budget, API rate limiting, loop detection
- Health Model: Automatic degradation and recovery
Configuration
- MAX_TOKENS_PER_MINUTE: 50,000
- MAX_TOOL_CALLS_PER_MINUTE: 60
- LOOP_SIMILARITY_THRESHOLD: 0.9
- TEMPO_COMPRESSION_RATIO: 0.3
Stats
- 31 tests passing
- TypeScript clean build