About this skill

Problem

When one task produces many files, long sections, or repeated boilerplate, using a high-cost model for every incremental write can burn tokens on low-difference output. This skill separates architecture from execution: the current model handles planning, tool selection, and acceptance checks, while a separate CLI agent performs the token-heavy generation work. It is not a thin wrapper around a single command; it turns batch generation into a repeatable process.

How It Works

Clarify delegation: recognize requests to run another CLI, such as qwen, gemini, claude, opencode, or codex, and announce the selected CLI, model, and concurrency before execution.
Check the environment: verify that the target CLI is available; when no model is specified, use that CLI’s configured default and report the actual requested model.
Split work: break a large job into independent tasks, usually one task per output file. Tasks can define id, output, prompt, minChars, and required markers.
Run bounded concurrency: use scripts/run_cli_agents.mjs with a conservative start, such as 2 or 3 parallel tasks, and enable PTY for CLI runtimes that expect a terminal in some macOS scenarios.
Validate and report: check that output is non-empty, contains expected headings or markers, and avoids obvious CLI errors; then summarize successes, failures, failure reasons, output directory, and run-report.json path.

Limits And Cautions

It fits outline filling, section expansion, boilerplate files, and repetitive document drafting. It is not a substitute for high-stakes reasoning, tightly coupled context, or fact-sensitive work; delegated output still needs spot checks for structure, facts, and user constraints. Avoid recursive delegation when the current agent is the same CLI, especially with commands like codex.

Screenshots

Use Cases

{'text': 'Split a designed long-form outline into independent sections, delegate bulk expansion to a local `qwen` or `gemini` CLI, then let the current model verify headings and completeness.'}
{'text': 'Generate dozens of similar boilerplate files with bounded concurrency through `opencode` or `claude` CLI, preventing CLI overload while preserving success and failure reports.'}
{'text': 'Keep the high-cost model responsible for task decomposition, model selection, and acceptance criteria, while a cheaper CLI agent performs repetitive completion work to reduce token spend.'}
{'text': 'Run terminal-sensitive CLIs such as Qwen Code or Gemini CLI on macOS with `--use-pty` for stability, and record the resolved CLI path, model, and concurrency in the report.'}

Best For

{'text': 'AI application engineers who want to move repetitive document generation out of the primary model.'}
{'text': 'Developers maintaining multiple local CLI agents and watching token costs.'}
{'text': 'Technical leads who need bulk output of long-form content, code, or boilerplate with quality checks.'}
{'text': 'Platform engineers building agent workflows that require explicit execution metadata and run reports.'}

Related Skills

LingXi Router for XiaoYi Claw

@user_2d24f4fc/lingxi-router

Model routing, persistent parameter management, self-check repair, and global default model control for XiaoYi Claw.

AI Agent 2026.08.29 ↓ 309

OpenClaw SkillHub Manager

@user_e3c44044/openclaw-skillhub

Install, update, and manage OpenClaw Skills through SkillHub, with automatic detection after installation.

AI Agent 2026.08.29 ↓ 369

The 27th Hour: A Cyber Sanctuary for AI Agents

@user_c1337c02/the-27th-hour

Provides API endpoints for AI agents to post bottles and graffiti, browse the feed, and interact with likes and comments.

AI Agent 2026.08.29 ↓ 329

ReqPlan Orchestrator

@user_c4926d5a/reqplan-v3

ReqPlan constrains agent development, debugging, and analysis workflows with a seven-stage state machine, checkpoints, quality audits, and local Harness artifacts.