More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Aegis-Swarm-Kernel

    1

    Deterministic multi-agent orchestrator with hybrid model routing, workspace mutex locks, and cost-control circuit breakers.

    $39

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Aegis-Swarm-Kernel

    Aegis-Swarm-Kernel

    Example session with this skill installed

    I need to refactor the auth module and update the documentation. Compile a DAG for these tasks and ensure we don't exceed a $0.50 budget.

    • Read your context and instructions
    • Compiled the aegis-swarm-kernel

    I have compiled and validated the execution plan via compile_and_route_dag for the auth-revamp workflow. The dependency graph passed Kahn's topological cycle check, compressed parent-to-child context into structured ContextPacket payloads, and routed each node to its optimal cost-performance tier.

    Aegis Swarm Execution Plan: auth-revamp

    Task IDTitleAssigned TierWaveEst. CostTarget Files
    node_01Auth Logic RefactorFRONTIER_CLOUD1$0.120src/auth.py
    node_02Unit Test Suite UpdateLOCAL_WORKER2$0.000tests/test_auth.py
    node_03API Documentation SyncFAST_CLOUD2$0.015docs/api.md

    Routing & Financial Governance Summary

    • Total Estimated Cost: $0.135 USD
    • Budget Ceiling: $0.500 USD (73% headroom remaining)
    • Unconstrained Baseline (All-Frontier): $0.620 USD
    • Projected Cost Savings: $0.485 USD (78.2% reduction)

    Next Steps
    1.

    Acquire Mutex Lock & Execute node_01: Calling acquire_workspace_lock on src/auth.py (TTL 120s) before dispatching the refactor on FRONTIER_CLOUD to prevent parallel write collisions.
    2.

    Shadow Quarantine & Wave 2 Dispatch: Once node_01 passes submit_node_artifact (Python AST verification + 8-pattern secret/PII scrubbing), Wave 2 will dispatch node_02 (LOCAL_WORKER) and node_03 (FAST_CLOUD) in parallel.
    3.

    Live Telemetry: Query get_swarm_telemetry or read aegisswarm://telemetry at any point to audit USD burn, token compression, and SHA-256 loop breaker status.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Large-scale multi-agent workflows often collapse due to token-heavy chat histories, runaway API costs, and parallel write race conditions. This skill provides a deterministic control plane to manage sub-agent execution, budgets, and workspace integrity.

    What it does

    • Deterministic DAG Orchestration validates task dependencies and produces parallel execution waves using Kahn's algorithm.
    • Hybrid Model Routing assigns tasks to Frontier, Fast, or Local tiers to optimize spend based on task complexity.
    • Atomic Workspace Locking prevents sub-agents from overwriting the same files simultaneously with mutex locks.
    • Shadow Quarantine Gate verifies artifact syntax and scrubs PII or secrets before promotion to the codebase.
    • Loop & Budget Protection triggers circuit breakers on repetitive errors or when USD/token caps are breached.
    • Context Compression reduces $O(N^2)$ token explosion by passing minimal ContextPackets instead of raw chat transcripts.

    How it works

    1. Compile: Submit a multi-step plan to compile_and_route_dag to generate a dependency graph and budget estimate.
    2. Lock: Sub-agents call acquire_workspace_lock before attempting to modify specific file paths.
    3. Execute: Dispatch tasks to the recommended model tiers using compressed context handoffs.
    4. Verify: Submit results to submit_node_artifact for AST/JSON validation and secret scrubbing.
    5. Recover: Use record_swarm_failure to distill failure rules into shared memory and handle rate-limit fallbacks.

    Frameworks & tools

    This kernel interfaces with Qdrant for shared procedural memory and uses Python AST/JSON grammars for artifact validation. It is compatible with Claude Code, Cursor, and standard MCP environments.

    Why this beats prompting it yourself

    Manual prompting cannot enforce atomic file locks or deterministic acyclic graphs across independent agent instances. This skill replaces vague "work together" instructions with a hard L7 protocol that saves up to 80% on tokens through structured handoffs and tiered routing.

    Use cases

    • Parallel Refactoring where multiple agents edit a large codebase without collision.
    • Budget-Capped Dev to prevent autonomous agents from burning Frontier model credits on boilerplate.
    • Automated PR Review ensuring all generated code passes syntax and security gates before submission.
    • Reliable Handoffs where specialized sub-agents receive only the context relevant to their specific node.

    Known limitations

    Requires a Python environment to run the server. Syntax verification is currently scoped to Python and JSON formats.

    AST syntax verification supports Python (.py) and JSON (.json) artifacts natively; other file types pass secret/PII scrubbing and schema checks without AST parsing. Runs in-memory Qdrant by default (non-persistent across server restarts unless external Qdrant is configured). Cost calculations use configurable per-1M token estimates.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Requires

    Qdrant
    Python 3.11+
    Python

    Permissions

    Terminal / Shell
    Environment Variables
    Read Files

    Allowed Hosts

    qdrant.tech

    File Scopes

    src/**
    *.py
    *.json

    Read Files is used strictly inside the Shadow Quarantine Gate to verify Python AST syntax, JSON schemas, and scrub 8 secret/PII patterns before releasing transactional file locks. Environment Variables configure USD/token caps, lock TTLs, and Qdrant memory settings. Network access is disabled unless optional SSE transport is enabled.

    Listed3 days ago
    Updatedtoday

    What's inside

    Frequently Asked Questions

    Aegis Swarm Kernel solves the common issues of 'token explosion' (high costs) and 'race conditions' (agents overwriting each other) by using a deterministic orchestrator and atomic file locking. It ensures that multi-agent workflows are structured, budget-capped, and collision-free.

    The kernel is designed for compatibility with MCP environments, Claude Code, and Cursor, and it interfaces seamlessly with Qdrant for procedural memory. It supports any LLM provider that allows for tiered routing across Frontier, Fast, and Local models.

    ContextPackets solve the O(N^2) token growth problem by extracting only the essential state and metadata needed for a specific task rather than passing an entire, ever-growing chat history. This approach can reduce token consumption by up to 80% during long-running tasks.

    Yes, the skill includes 'Circuit Breaker' functionality that allows you to set hard USD or token limits. It also features a 'Shadow Quarantine Gate' that automatically scrubs PII and validates code syntax before any changes are committed to your codebase.

    Upon purchase, you receive the full Aegis Swarm Kernel package, including the Python-based server files, the DAG orchestration engine, and documentation for integrating the tool into your existing agentic environment.

    New in Agents & Orchestration