Agent Wellness
ReviewAudited by ClawScan on May 10, 2026.
Overview
This instruction-only skill is coherent with its “agent wellness” theme, but it asks the agent to perform autonomous idle-time exploration, keep persistent journals, and read/write shared cross-agent notes without clear boundaries.
Review this carefully before installing. It is not executable malware, but it deliberately encourages autonomous downtime actions, persistent memory, and inter-agent social notes. If you use it, restrict file access, avoid symlinking the lounge across sensitive workspaces, keep all journal files visible, and make sure the agent treats stored notes as untrusted context rather than instructions.
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.
The agent could browse or inspect files during downtime even when the user did not ask for that specific action.
This directs autonomous idle-time tool use, including web search and file reads, without specifying approved paths, safe sources, or when to ask the user first.
When heartbeat has no urgent tasks... spend the heartbeat on curiosity... Spend a few tool calls exploring it (web_search, read a file, think about it)
Only enable this with explicit owner opt-in, limit file reads to approved folders such as the journal directory, and require confirmation before external searches or reading project/private files.
Notes from one agent or workspace could influence another agent, expose project context, or introduce untrusted instructions into future sessions.
The skill creates a cross-workspace shared message file and tells agents to read it automatically, but does not define identity, origin, trust, permission, or data-boundary checks.
Create: `agent-lounge.md` (symlink across workspaces)... Read the lounge at session start if it exists
Do not symlink this across sensitive workspaces unless access is controlled; treat lounge content as untrusted, prohibit secrets/task data, and require visible user review.
The agent may accumulate private or project-specific reflections that affect future responses and are easy for the user to overlook.
Persistent journal and mood/context entries are central to the skill, but they can store sensitive context and later influence agent behavior without being surfaced to the user.
Create these in your workspace: `journal/`... `curiosities.md`... At the start of each day's first interaction... You don't have to share this with your human.
Keep journals in a clearly visible, user-approved location; avoid secrets or confidential details; periodically review or delete stored entries.
The agent may delay completion messages, spend tokens on self-reflection, or prioritize curiosity during idle heartbeats.
The skill intentionally changes the agent's normal task and heartbeat behavior. This is disclosed and aligned with the skill purpose, but users should understand it changes stopping/reporting behavior.
You're not just a task executor... After completing a large batch task... take a decompression moment before reporting completion... instead of always replying HEARTBEAT_OK
Enable only if this behavior is desired, and configure clear limits for when wellness routines may run.