Zero-LLM one-shot reminders (<48h) via nohup sleep + openclaw message send, operated via {baseDir}/scripts/nohup-reminder.sh.
Security Analysis
high confidenceThe skill's code, runtime instructions, and declared requirements are coherent with its stated purpose (short-lived reminders via the openclaw CLI); nothing requested or installed is disproportionate or unexplained.
Name/description, required binaries (jq, openclaw), and the included script all match: the skill composes a message at creation time and uses a background nohup sleep + openclaw message send to deliver it. There are no unrelated credentials, binaries, or platform installs requested.
SKILL.md instructs the agent to compose messages, read TOOLS.md (and fallback to session_status.deliveryContext.to) and possibly write TOOLS.md when delivery target is missing. The included script itself only expects --target and does not touch other system credentials. Be aware this implies the agent will read and may write workspace files (TOOLS.md, reminders.json) — which is expected for delivery-target discovery but is a scope to note.
No install spec — instruction-only with a bundled script. No remote downloads or archive extraction; nothing is written to system locations beyond workspace files and /tmp, which is consistent with a lightweight helper.
The skill declares no secrets and only requires the openclaw CLI (which uses existing gateway credentials). The script accepts a few environment overrides (REMINDERS_JSON, REMINDER_TARGET, REMINDERS_LOCK_DIR) which are reasonable for configurability. Note: it will use whatever authentication the local openclaw binary is configured with, but it does not request or read tokens itself.
The skill does not set always:true. It spawns detached background processes (one per reminder) under the invoking user; reminders do not survive reboots and the script documents this limitation. Autonomous invocation is allowed (intentional for assistant workflows) — this is expected but you may choose to disable autonomous invocation if you want explicit control.
Guidance
This skill appears to do exactly what it says: short-horizon reminders delivered by the local openclaw CLI. Before installing, check these points: 1) The agent will read (and may write) workspace files: TOOLS.md (for delivery target) and reminders.json (tracking). If your workspace is shared or backed up, be comfortable with reminder metadata being stored there. 2) Deliveries use your local openclaw CLI configuration — ensure your OpenClaw gateway credentials and channels are properly secured. 3) Background reminders spawn detached sleep processes under the same user; they are lightweight but can accumulate if abused (monitor or limit usage if needed). 4) Reminders do not survive machine reboot. 5) If you do not want the agent to create reminders autonomously from natural conversation, set disable-model-invocation: true. Overall this skill is internally coherent and proportionate; use the above mitigations if any of the noted behaviors are unacceptable for your environment.
Latest Release
v1.1.4
- No user-visible changes; - README updated.
More by @lstpsche
Published by @lstpsche on ClawHub
