Task Resume

Automatic interrupted-task resume workflow with queueing and recovery. Use when a user asks to resume interrupted work after temporary context switches, prot...

This skill appears to implement what it says (a cross-session interrupted-task queue) but has several operational risks you should consider before installing: - Shared queue: It stores queued tasks in a workspace-global file (memory/task-resume-queue.json) shared across main/clone/group sessions. If you have multiple users or isolated contexts in the same workspace, tasks and their context could be visible across sessions. Avoid putting secrets in task context. - Log reading: The recover command can read arbitrary session .jsonl log files (e.g., under ~/.openclaw). Ensure you trust the skill and its invocations before allowing it to read local logs, since those logs may contain sensitive content. - Autonomous/cron auto-continue: The README and SKILL.md recommend a watchdog cron that automatically 'continues' tasks. That behavior can cause the agent to take actions or send progress messages automatically. If you need strict user confirmation for actions, do not enable the watchdog or restrict how/when the skill is invoked. - Operational controls: If you decide to install, restrict who/what can call the skill, avoid storing secrets in queue entries, and consider placing the workspace memory directory in a location with appropriate access controls. Review any automation (cron/heartbeat) separately and prefer manual invocation when in doubt. If you want me to, I can suggest a hardened variant (e.g., per-session queue files, explicit opt-in for watchdog mode, and an allowlist for recover log paths) to reduce these risks.

Type: OpenClaw Skill Name: task-resume Version: 1.3.0 The `scripts/task_resume_queue.py` script contains a significant information disclosure vulnerability. Its `recover` command uses `Path(log_path).expanduser()` to read the content of a specified log file. If an attacker can control the `--log` argument, this allows reading arbitrary files within the user's home directory (e.g., `~/.ssh/id_rsa`, `~/.aws/credentials`). The last 8 lines of the read file are then stored in the `context` field of `memory/task-resume-queue.json`, making sensitive data accessible to the agent for future processing. Additionally, the `SKILL.md` instructs the agent to 'resume' tasks based on stored `title`, `context`, and `acceptance` fields, which could lead to prompt injection against the agent if these fields are manipulated.