Shark

Enables non-blocking AI agent execution by spawning parallel remora subagents for slow tasks, keeping the main agent responsive and efficient.

What this skill will actually do: it spawns background jobs, writes and reads state files (pending.json, timings.jsonl, .shark-done, SHARK_LOG.md), schedules a periodic poller/cron job, and may run arbitrary shell commands (gh, ssh, build/test commands, or launch an LLM runtime). These capabilities are coherent with 'non-blocking execution' but mean the skill can execute anything your agent/runtime user can — including commands that access SSH keys, GitHub tokens, or other workspace secrets. Before installing: (1) review the included scripts (shark.sh and poll-and-deliver.js) to ensure they do only what you expect; (2) run the skill in an isolated environment or a sandboxed agent first; (3) restrict runtime permissions if possible (avoid giving the skill access to secrets or global cron unless necessary); (4) be cautious about the calls that use 'powershell -ExecutionPolicy Bypass' and 'claude --permission-mode bypassPermissions' since those flags relax protections; (5) if you install, periodically check pending.json and any created cron jobs and remove them when not needed.

Type: OpenClaw Skill Name: shark Version: 0.1.0 The 'shark' skill implements a non-blocking execution framework for AI agents, referred to as the 'Shark Pattern.' It provides instructions and scripts (shark.sh, shark-exec/SKILL.md) to decompose slow tasks into parallel sub-agents ('remoras') or background processes, ensuring the main agent turn remains responsive (under 30 seconds). The skill utilizes standard OpenClaw features such as sessions_spawn, cron-based polling, and background execution to manage these tasks. While the implementation involves complex agent instructions for state management and process monitoring, the behavior is entirely consistent with the stated goal of optimizing agent performance and lacks any indicators of malicious intent or unauthorized data access.