Windowed Autonomy
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill is openly an autonomy scheduler, but it also gives the agent broad self-directed work authority, persistent memory influence, and a hard-coded MONEY goal that users should review before installing.
Use this only if you intentionally want scheduled autonomy. Before enabling it, remove the hard-coded MONEY goal, define exactly which tasks and actions are allowed, require approval for high-impact changes, and review the memory/checkpoint files the agent will write and later rely on.
Findings (5)
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.
The agent may prioritize or record work around a money-focused goal even if that is not what the user wants.
A generic autonomy skill hard-codes a specific long-term goal and asks the agent to update the user's goals file, which can redirect future work away from the user's actual intent.
Every task should link to RA's long-term goal: **MONEY** ... After completion: Update GOALS.md with progress notes
Remove the hard-coded MONEY goal, make goals explicitly user-defined, and require confirmation before modifying GOALS.md.
If the queue contains risky or ambiguous tasks, the agent could act without asking first during heartbeat sessions.
The instructions authorize the agent to autonomously select and complete queued tasks, including urgent work outside the normal time window, without specifying action limits or approval requirements for high-impact tasks.
Pick the first task you can work on ... Work on task until: Complete ... URGENT found: Work immediately (override window)
Add explicit guardrails requiring user approval for deletion, publishing, purchases, account changes, external communications, credential use, or production changes.
Incorrect or poisoned memory notes could cause the agent to resume the wrong work, follow unsafe instructions, or preserve sensitive goals/preferences longer than intended.
The skill stores persistent memory and later makes stored resume instructions authoritative, creating a path for stale, mistaken, or modified notes to steer future autonomous behavior.
Write checkpoint to memory/episodic/[today].md IMMEDIATELY when context ≥70% ... Checkpoints include "Resume Instructions" — follow them exactly
Treat checkpoints as advisory, validate them against the current user request, exclude sensitive data by default, and avoid 'follow exactly' language for stored memory.
Installing or integrating this skill may cause the agent to act during configured heartbeat windows instead of waiting for direct prompts.
The recurring autonomous behavior is the stated purpose and is disclosed, but it changes the agent from reactive use to scheduled self-directed work.
The agent works autonomously only during defined time windows ... Daytime (8 AM - 8 PM UTC): Autonomy ON, work from queue
Use only with a queue you control, narrow the active windows, and monitor the first several autonomous sessions.
Users have less provenance information for deciding whether this published instruction set is trustworthy.
The package is instruction-only, so there is no executable dependency risk in the supplied artifacts, but the metadata does not provide a verifiable source commit.
"commit": "https://github.com/openclaw/skills/commit/placeholder"
Publish a real repository or commit reference and verify the skill text before integrating it into an autonomous workflow.