Agent Governance Middleware
YAMTAM
Yet Another Multi-Tier Agent Middleware · ENGINE
v0.12.0 · release 2026-05-28 · 826 checks · PASS
87Agents
156Commands
27Hooks
350Skills
58Rules
35Scripts
Hook Execution Pipeline — 27 hooks · mỗi tool call đi qua các lớp này theo thứ tự
UserPrompt
session-bootstrap
→
PreToolUse · blocks here
token-budget-guard
guard-destructive
scope-guard
api-destruct-guard
db-protect
deploy-gate
commit-gate
code-freeze
cost-guard
context-gate
rbac-guard
token-scope-guard
→
AI Executes
TOOL CALL
permission-auto-approve
tool-attention.js
context-monitor.js
gitnexus-hook.js
→
PostToolUse · logs here
audit-log
context-gate-log
format-on-write
telemetry-sender
log-agent
auto-qa-trigger
auto-qa-reset
auto-kill-stuck-tasks
→
Stop · after turn
truth-gate-guard
validate-completion
Agent Registry — 87 agents · 11 chuyên môn · gọi bằng @agent-name
⚙Core Dev
  • fullstack-engineer
  • api-designer
  • api-gateway-engineer
  • event-driven-architect
  • graphql-architect
  • microservices-architect
  • monorepo-architect
  • websocket-engineer
8 agents
🏗Infrastructure
  • devops-engineer
  • kubernetes-specialist
  • terraform-engineer
  • cloud-architect
  • sre-engineer
  • platform-engineer
  • deployment-engineer
  • incident-responder
  • security-engineer
  • network-engineer
  • performance-engineer
  • database-admin
12 agents
🧪Quality & Test
  • qa-engineer
  • test-architect
  • qa-automation
  • chaos-engineer
  • compliance-auditor
  • accessibility-specialist
  • penetration-tester
  • code-reviewer
  • debugger · qa-expert
  • test-automator
  • architect-review
11 agents
🧠Data & AI
  • llm-architect
  • ai-engineer
  • ml-engineer
  • data-engineer
  • data-scientist
  • nlp-engineer
6 agents
📊Business
  • business-analyst
  • scrum-master
  • technical-writer
  • ux-researcher
  • project-manager
  • strategic-analyst
  • release-manager
7 agents
🎼Orchestration
  • task-orchestrator
  • multi-agent-coordinator
  • workflow-director
  • knowledge-synthesizer
  • task-decomposer
  • planner · spec-planner
  • spec-executor
  • spec-verifier
9 agents
🛠Dev Experience
  • build-engineer
  • git-workflow-manager
  • mcp-developer
  • refactoring-specialist
4 agents
🛡Security
  • security-auditor
  • prompt-firewall
  • code-auditor
  • architecture-auditor
  • performance-auditor
  • dependency-analyzer
6 agents
⚡Standalone
  • frontend-developer
  • backend-developer
  • database-expert
  • docker-expert
  • cicd-engineer
  • systems-architect
  • project-architect
  • ui-ux-designer
  • documentation-writer
  • token-guard · tool-router
32 agents
Rule Registry — 58 rules · Laws + Guards + Policies
immutable-infrastructure-law financial-deadman-switch-law sovereign-runtime-law secrets-vault-law network-egress-whitelist-law bft-consensus-law circuit-breaker-law canary-deployment-law dependency-sandbox-law token-budget-velocity-law code-signing-law sovereign-overlord-gate-law agent-excessive-agency-law agent-hierarchy-law agent-middleware-law anti-evasion-law memory-persistence-law network-egress-law resource-quota-law shell-sanitize-law slsa-artifact-law observability-telemetry-law prompt-jailbreak-advanced supply-chain-vetting honeypot-trap-law agent-tool-poisoning-guard prompt-jailbreak-guard fuzz-testing-constraints privilege-isolation sandbox-isolation-law container-hardening-law human-gate-policy token-budget-policy audit-hardening-policy agent-communication-policy env-integrity-policy rule-consistency-policy subagent-policy conflict-resolution git-push-enforcement git-workflow-v2 owasp-llm-output-law dependency-vetting-law api-security-gate meta-rule-enforcer terminal-validator agent-code-constraints execution-environment verification testing · tests golden-principles security typescript migrations agents-v2 color-rules typography-rules
Skill Domains — 350 skills · knowledge chunks Claude dùng khi cần
AI / LLM / Agents 68
react-agent-loop rewoo-plan-execute reflexion-verbal-rl memgpt-virtual-context tree-of-thoughts multi-agent-debate eval-driven-agent-dev agentic-failure-modes
Security / Crypto 45
ecc-key-management jwt-jws-jwe owasp-llm-top10 zero-trust-patterns red-team-check supply-chain-security honeypot-deception
ML Infra / Inference 32
vllm-paged-attention flash-attention llama-cpp-quantization triton-inference deepspeed-zero kv-cache-optimization
K8s / Cloud Native 28
k8s-crd-controller argocd-gitops istio-traffic crossplane helm-chart linkerd-mesh
Distributed Systems 28
raft-consensus etcd-config crdt-sync grpc-streaming protobuf snowflake-id
UI / Design / Frontend 42
design-tokens-system aesthetic-anchor motion-physics typography-scale accessibility-audit shadcn-patterns
Observability / Ops 25
opentelemetry prometheus jaeger-tracing loki-log slo-design incident-runbook
Linux / Sandbox / Kernel 22
ebpf-syscall-monitoring runtime-sandbox-runc overlayfs-isolation seccomp-filtering bubblewrap capability-dropping
// Memory Architecture
L1
Atomic Facts — Persistent
Project truths · Sovereign identity · Skills lock (350) · Known constraints · Git-tracked · Confidence-gated
L2
Session Facts — Ephemeral
Token budget · Circuit state · Scope declaration · Trust score · Checkpoint index · Git-ignored
L3
Vector Search — Out of scope
Semantic retrieval · Not implemented · See .out-of-scope/
// Safety Gate Stack
L3 STOP
truth-gate-guard.sh
Scan final turn for claim verbs without evidence. Non-blocking warn → self-correct.
L2 PRE
token-budget-guard.sh
Circuit Breaker CLOSED→OPEN→HALF-OPEN. HARD BLOCK at 5 consecutive calls.
L2 PRE
guard-destructive.sh
Intercept rm/drop/delete/truncate. Require explicit confirmation for destructive ops.
L1 POST
audit-log.sh
Hash-chain JSONL of every tool call. SHA-256 prev→current. Tamper-evident.
L0 USER
session-bootstrap.sh
Inject matching L1 facts + trust score + budget mode into every prompt context.
YAMTAM ENGINE v0.12.0 · release 2026-05-23 · 477 checks · 24/24 PASS · stateless · no daemon · no network required
Claude Code Cursor Aider VS Code MIT License