meerkat: Skill for Claude Code

Meerkat Platform Guide

Meerkat (rkat) is a library-first agent runtime exposed through multiple surfaces. One execution pipeline, realm-scoped state, runtime-backed semantics by default. This guide is task-oriented; deeper schemas live under references/.

Where to find details

This guide is an overview. The reference documents below carry the detail; read the matching one before answering a detailed question. On a help surface (rkat help, JSON-RPC help/ask, REST POST /help, MCP meerkat_help), read a reference with the skill_read_resource tool, passing this skill's source_uuid and skill_name and the reference path as path; skill_list_resources lists them.

• references/api_reference.md - per-surface methods, schemas, examples (CLI, REST, RPC, MCP, Python/TS/Rust SDKs).
• references/sessions_runtime_surfaces_sessions.md - sessions on every surface: create, resume, list, read, history, archive, turn queueing and the attempt cap, interrupt versus turn/stop_run, forks, transcript edits, system prompts, events and replay, ATIF export, durable runs (0.9.5 per-round commits, same-run resume, tool reruns), durable resume holds and tool-process recovery.
• references/sessions_runtime_surfaces_storage_config.md - realm ids and where state lives, backends, host stores (0.9.5), multiprocess hosting, SQLite, rkat storage, runtime health and shutdown, WorkGraph limits, schedules, rkat config and config keys.
• references/sessions_runtime_surfaces_api_changes.md - JSON-RPC methods and REST routes added or renamed since 0.8.40, the ErrorCode table with HTTP statuses and rkat exit codes, SDK changes, binaries and installs, Rust embedding, removals.
• references/providers_and_auth_models.md - choosing a provider and model, default models and the catalog, capabilities, thinking and effort, provider_params and prompt caching, realtime, voice and image models, self-hosted servers, provider errors, refusals and content filters.
• references/providers_and_auth_credentials.md - env-var API keys, realm backends, auth profiles and bindings, base URLs and cloud routes, credential storage and keychain refusals, every rkat auth subcommand, --auth-binding, auth RPC/REST/SDK methods and errors, connector OAuth clients.
• references/security.md - OS sandboxing (process confinement) and what is not a sandbox, shell command policy, tool access policy and read-only, where credentials and secrets live, MCP security boundaries, audit, security errors and refusals.
• references/security_governance.md - governed realms (rkat realm govern, grants, --as-agent, governed rkat-rpc), review tiers R1/R2/R3, human consent and approvals, native authorization for Rust hosts, authorization changes while work runs.
• references/mcp_tools_skills_mcp.md - MCP servers: mcp.toml and realm config, every rkat mcp flag, which hosts start configured servers, MCP OAuth, access ceilings, live mcp/*, deferred tools, MCP Apps, MCP errors.
• references/mcp_tools_skills_tools.md - built-in tool families and exact names, --tools presets, [shell]/[tools] config, tool errors, filters and overlays, tool_access_policy, streaming tool deadlines, callback tools.
• references/mcp_tools_skills_skills.md - skills: SKILL.md format, identity and sources, [skills] config, rkat skill, rkat run --skill, preload_skills and skill_refs, mob skills, introspection, skill errors.
• references/hooks_models_hooks.md - hook points, [[hooks.entries]], command/http runtimes, hooks_override, 0.9 mandatory model-request hooks, hook events.
• references/hooks_models_model_operations.md - max_tokens, budgets and turn limits, compaction, structured output, retries and timeouts, model fallback and model switching (prompt caching: references/providers_and_auth_models.md).
• references/hooks_models_turn_control.md - interrupt, turn/stop_run, request cancel, parallel tool calls, event types and delivery.
• references/live_comms_mobs_live.md - live (realtime voice) channels: live/* on rkat-rpc, realtime models, audio and size limits, SDK calls, the GPT Live preview, mob member live (rkat mob live).
• references/live_comms_mobs_comms.md - agent-to-agent comms: transports, [comms] config, rkat run --comms-* flags, comms tools, comms/send on every surface and its delivery outcomes, remote signed peers.
• references/mobs.md - multi-agent orchestration in depth.
• references/live_comms_mobs_mobs.md - mob changes since 0.8.40 (through 0.9.4): stop, shutdown, destroy and admission close, member lifetime, member binding history, helpers, fork_off, council, profile keys, flows and runs, multi-host mobs, mobpacks and deploy, error codes.
• references/migration_0_9.md - upgrading a 0.8 system to 0.9 (renames, config that no longer loads, behaviour changes), and 0.9.3 to 0.9.4.
• references/migration_0_5.md - only if you're upgrading from 0.5 (host_mode → keep_alive, etc.).