claude-afklisted

# claude-afk — Observable Headless `claude` ## When to use You want the shape of `claude -p --output-format json` (give prompt, block, get JSON, exit), but you also want to *see what the agent is doing* while it runs — not just the final JSON. Applies when: - You're orchestrating one-shot agentic invocations from scripts, cron, or other agents. - You want to be able to `tmux attach` mid-run to watch tool use, intervene, or debug. - You want the full session transcript available after exit, not just the summarized `result`. Primary value is **transparency and observability**: a real Claude Code session under the hood means you get the same UI, the same tool-call rendering, the same scrollback, and the same hook surface as if you'd typed the prompt yourself. (Side benefit: because the work runs inside an interactive Claude Code session, it bills against the Max plan's Claude-Code bucket rather than the extra-usage/API bucket — useful if you're on Max with extra-usage off. Not the reason to reach for this skill, just a happy side effect.) ## Contract (what callers can rely on) Input: ``` claude-afk <prompt> # positional, required [--system-prompt <str>] # optional system prompt [--out <path>] # JSON output path; default: mktemp [--timeout <seconds>] # kill session after N sec; default: 1800 [--session-prefix <str>] # tmux session name