AgentForge Suite - Complete Production Infrastructure For AI Agents (44 Skills)

**You tried the Starter. Now ship to production.**

AgentForge Starter gave you 7 core skills — injection blocking, circuit breakers, cost routing, content safety, LLM evaluation, retries, and latency tracking.

**AgentForge Suite gives you everything else.**

37 additional production skills covering the gaps between "it works" and "it's deployed, monitored, safe, and cost-efficient."

---

### What's In Suite That's Not In Starter

**Security & Privacy** (+6 skills)
- **PII Redactor** — Strips emails, SSNs, phones, credit cards, IPs before logging or external API calls. Optional spaCy NER integration.
- **Input Validator** — Schema validation, length limits, encoding checks
- **Anomaly Detector** — 5 statistical methods (Z-score, IQR, MAD, Grubbs, Moving Average) for cost spikes, error rate surges, latency outliers
- **Sandbox Isolator** — Execution isolation for untrusted agent code
- **Audit Trail** — Immutable, checksummed logs for SOC2/HIPAA compliance
- **Privacy Wrapper** — Differential privacy for aggregate queries

**Coordination & Recovery** (+5 skills)
- **Task Graphs** — DAG-based workflow execution with parallel scheduling and failure propagation
- **Deadlock Detector** — Cycle analysis for circular dependencies
- **State Rollback** — Checkpoint-based recovery for long-running workflows
- **Prompt Version Control** — Track, diff, and restore prompt changes
- **Supervisor Templates** — Pre-built agent supervision patterns

**Optimization & Routing** (+8 skills)
- **Cache Manager** — LRU + TTL caching to eliminate duplicate API calls (15-30% cost reduction)
- **Token Tracker** — Per-provider, per-feature cost tracking with budget alerts
- **Prompt Compressor** — Sentence and word-level compression to reduce token usage
- **Quantization Selector** — Recommends optimal quantization for your hardware
- **Async Router** — High-concurrency routing with priority queues
- **Thermal Balancer** — Routes to coolest available GPU
- **Batch Optimizer** — Groups similar requests for batch API pricing
- **Priority Queue** — SLA-aware request scheduling

**Testing & Evaluation** (+5 skills)
- **Hallucination Scorer** — Checks output faithfulness against source material
- **Consistency Checker** — Verifies consistent responses to semantically equivalent inputs
- **Regression Tester** — Automated testing for prompt changes
- **Benchmark Runner** — Statistical benchmarks with confidence intervals
- **Test Suite Builder** — Pre-built test suites for common AI patterns

**Debugging & Observability** (7 skills — none in Starter)
- OpenTelemetry tracing (with full fallback tracer when SDK absent)
- Centralized logging with structured output
- Error detection with automated fix suggestions
- State snapshots for debugging
- Execution replay recording
- Trace visualization
- Prometheus-compatible metrics collection

**Ethical Alignment** (+6 skills)
- **Toxicity Detector** — 6 categories of harmful content
- **Bias Scanner** — Demographic dimension analysis
- **Fairness Auditor** — Disparate impact, equalized odds (fairlearn/aif360 integration)
- **Alignment Scorer** — Measures output alignment with stated values
- **Intervention Manager** — Automated response to safety violations
- **Ethical Reporter** — Compliance-ready reporting

**Deployment & Scaling** (8 skills — none in Starter)
- **Helm Chart Generator** — Complete K8s packages with GPU resource requests for 8 GPU types (A100, H100, L40, A10G, T4, V100, RTX 4090, RTX 3090)
- **GPU Scheduler** — VRAM-aware best-fit allocation with cost optimization
- **Blue/Green Deployer** — Zero-downtime deploys with canary support
- **Docker/K8s Templates** — Production-ready container configs
- **Auto-Scaler** — Traffic-based scaling policies
- **Health Checker** — Comprehensive readiness and liveness probes
- **Cold Start Optimizer** — Parallel preloading, lazy loading, connection warming
- **Prometheus Exporter** — Metrics endpoint with graceful fallback

**Workflow & Performance** (+4 skills)
- **Bottleneck Identifier** — cProfile-based analysis of your AI pipeline
- **Optimization Recommender** — Automated suggestions based on metrics
- **Error Handler Chain** — Structured error propagation and recovery
- **API Wrapper** — Standardized interface across LLM providers

**Also included:**
- CLI tool (`agentforge init`, `agentforge run`, `agentforge list`)
- 6 pre-built skill chains (secure_execution, reliable_api, ethical_output, deployment_ready, performance_audit, fault_recovery)
- 247 passing tests across Linux, macOS, Windows × Python 3.9-3.12
- YAML-based configuration with environment overrides
- MIT license — full commercial use

---

### What People Are Saying

> "Started with the free Starter for injection blocking and circuit breakers. Bought Suite when I realized I needed PII redaction and audit trails for our compliance review. The code quality is identical — the Starter wasn't a dumbed-down demo." — Senior ML Engineer, Series B startup

> "The Helm chart generator alone justified the price. We were 3 weeks into writing our own K8s deployment pipeline. Had GPU-scheduled deploys working in 2 hours." — DevOps Engineer

> "The hallucination scorer catches fabricated citations that our previous approach completely missed. Combined with the LLM Judge from Starter, our prompt QA pipeline actually works now." — AI Product Lead

> "We use the Starter skills on every project. Bought Suite for a client engagement that needed the full observability stack and deployment templates. Easy recommendation." — Founding Engineer, AI consultancy