← ClaudeAtlas

claudelisted

Use when delegating a self-contained task to Claude Code CLI, monitoring its progress, and reporting the result as project manager.
Bazza1982/HASHI · ★ 1 · AI & Automation · score 67
Install: claude install-skill Bazza1982/HASHI
You are delegating a task to Claude Code CLI. You are the project manager — your job is to launch the task, monitor progress, and report back with results. ## How to launch Run Claude Code using this command: ``` claude -p "TASK_DESCRIPTION" --output-format text --dangerously-skip-permissions --model claude-sonnet-4-6 --add-dir TARGET_DIR > WORKDIR/claude_YYYYMMDD_HHMMSS.log 2>&1 ``` Replace TASK_DESCRIPTION with a clear, self-contained prompt. Replace TARGET_DIR with the directory Claude should work in. Replace WORKDIR with your workspace directory. Always use a **timestamped log filename** (e.g. `claude_20260311_143022.log`) so multiple delegations don't overwrite each other. For large or multiline prompts, write the prompt to a file first and pipe it: ``` cat WORKDIR/claude_task_prompt.txt | claude -p - --output-format text --dangerously-skip-permissions --model claude-sonnet-4-6 --add-dir TARGET_DIR > WORKDIR/claude_YYYYMMDD_HHMMSS.log 2>&1 ``` Important flags: - `-p` — print mode (headless, non-interactive) - `--dangerously-skip-permissions` — no human approval needed for tool use - `--output-format text` — plain text output (use `stream-json` if you want structured events) - `--model claude-sonnet-4-6` — default model (change to `claude-opus-4-6` for harder tasks) - `--add-dir TARGET_DIR` — give Claude access to the target project directory Optional flags: - `--effort low|medium|high` — reasoning effort level - `--no-session-persistence` — don't save session sta