Agents

Sandboxed skills.

The WASI 0.2 component runtime for executing community-supplied agent skills, MCP tools, and A2A skill components in a deterministic, capability-based sandbox. Language-agnostic. Content-addressable. Audit-ready.

STATUS: Testnet
CRATE: tenzro-wasm
STABILITY: Stable
TYPE: Runtime

Component manifest

{
  "id": "skill.tenzro.summarize",
  "version": "0.1.0",
  "content_hash_hex": "...",
  "runtime": "agent-skill",
  "capabilities": {
    "storage": { "read_only": [], "read_write": [] },
    "network": { "outbound_tcp": [], "https_proxy": false },
    "env": [],
    "host_methods": ["events.publish", "metrics.observe"]
  },
  "deadline_ms": 5000,
  "fuel_limit": 50000000
}

Defaults

Every capability list defaults to deny. A component with an empty manifest cannot read the filesystem, cannot open a socket, cannot read environment variables. Hosts grant capabilities explicitly.

Determinism

Wasmtime fuel metering counts WASM operations rather than wall-clock time. Two executions of the same component with the same input return the same fuel consumption regardless of host CPU speed.

Execution receipt

ExecutionReceipt {
  component_id, content_hash_hex,
  function, input_hash_hex, output_hash_hex,
  outcome: Success | Trapped | FuelExhausted | DeadlineExceeded | HostContractViolation,
  fuel: { budget, consumed, remaining, elapsed },
  completed_at_ms,
}

Embed points

tenzro-agent-kit (skill runtime, behind--features wasi-skills) and tenzro-node::mcp (sandboxed MCP tool host, same feature flag).

← All docs