任务执行工具

Sleep Task Executor MCP

This MCP server provides tools to generate core prompt + progress/index, run a single execution, or start continuous/interval loops via agent.

Tools

• sleep_task_execute

  • Inputs:
    • goal (string, required)
    • input_materials (string[], required)
    • output_dir (string, required)
    • workspace_dir (string, optional)
    • agent_bin (string, optional, default: agent)
    • run_once (boolean, optional, default: true)
    • show_generated (boolean, optional, default: true)

• sleep_task_start

  • Inputs:
    • goal (string, required)
    • input_materials (string[], required)
    • output_dir (string, required)
    • workspace_dir (string, optional)
    • agent_bin (string, optional, default: agent)
    • mode (string, required: continuous | interval)
    • interval_seconds (number, optional: continuous default 0, interval default 1800)
    • max_success_runs (number, optional: 0 or omitted = unlimited)

• sleep_task_list

  • Inputs:
    • status (string, optional: running | stopped | stopping | completed | error | stale)

• sleep_task_stop

  • Inputs:
    • task_id (string, required)

• sleep_task_log

  • Inputs:
    • task_id (string, required)
    • tail_lines (number, optional, default: 200, 0 = all)

Notes

• Requires agent CLI to be installed and logged in.
• Writes only to the specified output_dir.
• Uses fixed model: gpt-5.2-codex-xhigh-fast.
• Task database: data/tasks.json.
• Task logs: logs/<task_id>.log.

Run Semantics

• continuous: runs back-to-back; if interval_seconds > 0, waits after each run.
• interval: waits interval_seconds after each run (default 1800 seconds).
• max_success_runs: counts only successful runs; stop when reached.

Logging Details

• Each task logs tool invocation, run start/end, and wait intervals.
• Agent stdout/stderr are captured (truncated) per run.
• Output directory file changes (created/updated/deleted) are logged per run.