Claude Task Runner
WarnAudited by ClawScan on May 10, 2026.
Overview
The skill has a coherent headless Claude task-runner purpose, but it asks the agent to run an unreviewed local wrapper and recommends permission-bypassing Bash/write execution.
Use this skill only if you intentionally want unattended Claude Code execution on local files. Before installing, verify the referenced `cc-task-runner.sh` yourself, keep runs inside a limited workspace, avoid `bypassPermissions` unless necessary, and clean stored task state when prompts or outputs are sensitive.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A task run with this mode could execute shell commands or modify files more freely than a user may expect.
The skill recommends bypassing normal Claude Code permissions specifically for Bash and file-writing tasks, which can allow high-impact local changes without the usual interactive approval step.
**Permission mode**: Prefer `--permission-mode bypassPermissions` in trusted local environments when the task must write files or run Bash.
Use this only in a trusted, throwaway or well-scoped workspace; prefer approval-based modes unless the user explicitly requests unattended execution.
Installing the skill could cause the agent to execute whatever exists at that local path, or fail unpredictably if the path is absent.
The main runner script that the skill tells the agent to execute is not included among the reviewed package files and has no install specification, so its behavior and provenance cannot be verified from the supplied artifacts.
Primary script: - `/Users/zhengweidong/.openclaw/workspace/scripts/cc-task-runner.sh`
Package and review the runner script, avoid hard-coded user-specific paths, and declare required binaries and setup steps.
A task may continue running outside the immediate chat turn and may need explicit status, kill, or clean commands.
The runner is designed for background or long-running subprocesses and includes cleanup controls, which is purpose-aligned but still important persistence behavior.
Sub-process cleanup relies on pid file; if the host restarts, stale pids show as "running" until `kill` or `clean`
Check `status` and `list` after use, and run `kill` or `clean` for tasks that should not keep running.
Prompts and generated outputs may remain on disk after the task finishes.
The skill stores prompts, outputs, logs, and metadata persistently, which is disclosed and useful for task tracking but may retain sensitive task content.
All task metadata lives under `~/.openclaw/state/cc-tasks/<task-name>/`: ... `result.json` — raw Claude output (JSON) ... `prompt.txt` — the prompt that was sent
Avoid sending secrets in prompts and periodically clean task state if it contains private information.