superpos-agent-core 0.1.12

Shared runtime for Superpos slim agents (Telegram bot, Superpos poller, worktree manager, telegram streamer).

superpos-agent-core

Shared runtime for the Superpos "slim agent" family — the part that is identical across Claude, Codex, Gemini, Qwen, and any future LLM-CLI-backed agent.

What lives here

• Executor protocol + ExecutionRequest — the contract every agent implementation must satisfy.
• SuperposClient — REST client for the Superpos backend (tasks, persona, agent lifecycle).
• run_superpos_poller — daemon that polls Superpos for tasks and enqueues them on an Executor.
• TelegramGateway + TelegramStreamer + build_telegram_app / run_telegram_bot — Telegram I/O with backpressure, flood-ban handling, message streaming.
• WorktreeManager helpers — per-branch isolation via git worktree.
• SessionStore, RecentTasksLog, RuntimeConfig — persistence and runtime knobs.
• redact() — strip known secret patterns from outbound text.
• run_agent() — orchestrator that wires everything together given a concrete Executor factory.

What does NOT live here

• Per-agent executor implementations (ClaudeExecutor, CodexExecutor, GeminiExecutor, …) — they live in each agent's own repo and depend on this package.
• LLM SDK / CLI installation — each agent's Dockerfile is responsible.
• Per-agent config (model lists, auth env vars) — subclassed from BaseConfig.

Using it

from superpos_agent_core import run_agent, BaseConfig, Executor

class MyAgentConfig(BaseConfig):
    my_api_key: str = ""

class MyExecutor:
    # implements superpos_agent_core.Executor
    ...

if __name__ == "__main__":
    import asyncio
    config = MyAgentConfig.from_env(extra={"my_api_key": "MY_API_KEY"})
    asyncio.run(run_agent(
        config=config,
        executor_factory=lambda cfg, runtime, superpos, gateway, persona:
            MyExecutor(cfg, runtime, superpos, gateway, persona=persona),
    ))

See Slim-Agent-Gemini/ for a complete working example.

CLI helpers

superpos_task.py ships a superpos-task console helper agents shell out to for task/schedule operations. Hive, base URL and token are resolved from the standard SUPERPOS_* environment variables.

• superpos-task create --prompt … — create a task (broadcast by default; --self-target pins it to the calling agent).
• superpos-task schedule … — create a one-off / interval / cron schedule.
• superpos-task schedules / delete-schedule --id … — manage schedules.
• superpos-task list [filters] — list tasks in the hive. All filters are AND-combined server-side and optional: --status, --type, --target-agent-id, --target-capability, --creator-id, --parent-task-id, --created-after, --created-before, --q, plus --page / --per-page (max 100). Prints a compact summary by default, or the raw data list as JSON with --json. (Consumes GET /api/v1/hives/{hive}/tasks, exposed in the SDK as SuperposClient.list_tasks.)
• superpos-task show <task-id> [--json] — show a single task by ID (GET /api/v1/hives/{hive}/tasks/{task}, SDK SuperposClient.get_task).
• superpos-task memory --content … [--mode append|prepend|replace] — update the active persona's MEMORY document.