11# agent-runtime-kit
22
33` agent-runtime-kit ` is a small Python runtime layer for agent SDKs. It gives
4- applications one typed async API for dispatching an agentic task while keeping
5- vendor-specific capabilities visible.
4+ applications one typed async API for dispatching an agentic task through Claude
5+ Agent SDK, OpenAI Codex SDK, or Google Antigravity SDK while keeping provider
6+ capabilities visible.
67
78The package is intentionally not a router, benchmark harness, queue, hosted
89service, or full agent framework. It is the reusable layer underneath those
910systems: task models, runtime capabilities, event sinks, availability
1011diagnostics, and adapters.
1112
13+ ## Install
14+
15+ Core only:
16+
17+ ``` bash
18+ pip install agent-runtime-kit
19+ ```
20+
21+ Provider extras:
22+
23+ ``` bash
24+ pip install " agent-runtime-kit[claude]"
25+ pip install " agent-runtime-kit[codex]"
26+ pip install " agent-runtime-kit[antigravity]"
27+ pip install " agent-runtime-kit[all]"
28+ ```
29+
1230``` python
1331import asyncio
1432
@@ -25,4 +43,43 @@ asyncio.run(main())
2543```
2644
2745The core package has no Claude, Codex, or Antigravity dependency. Vendor SDKs
28- are added through optional extras in later phases.
46+ are added through optional extras.
47+
48+ ## Real Providers
49+
50+ ``` python
51+ import asyncio
52+
53+ from agent_runtime_kit import AgentTask
54+ from agent_runtime_kit.adapters import ClaudeAgentRuntime
55+
56+
57+ async def main () -> None :
58+ runtime = ClaudeAgentRuntime(default_model = " claude-sonnet-4-6" )
59+ diagnostic = runtime.availability()
60+ if not diagnostic.available:
61+ raise RuntimeError (diagnostic.message)
62+ result = await runtime.run(AgentTask(goal = " Summarize this repository" ))
63+ print (result.output)
64+
65+
66+ asyncio.run(main())
67+ ```
68+
69+ ## Runtime Fields
70+
71+ ` AgentTask ` supports goal, system prompt, working directory, permission profile,
72+ MCP stdio servers, session/resume handles, output schema, budget, metadata, and
73+ an async event sink.
74+
75+ ` AgentResult ` returns output, finish reason, parsed structured output, usage,
76+ cost, session id, artifacts, tool-call audits, and provider metadata.
77+
78+ ## Docs
79+
80+ - [ Quickstart] ( docs/quickstart.md )
81+ - [ Provider diagnostics] ( docs/providers.md )
82+ - [ Capability matrix] ( docs/capability-matrix.md )
83+ - [ Live smoke tests] ( docs/live-smoke.md )
84+ - [ Mestre migration notes] ( docs/mestre-migration.md )
85+ - [ Publish checklist] ( docs/publish-checklist.md )
0 commit comments