Agent
PassAudited by VirusTotal on May 12, 2026.
Overview
Type: OpenClaw Skill Name: agent Version: 1.0.0 The skill bundle consists entirely of markdown files defining an AI agent's identity, personality, voice, boundaries, and adaptation rules, along with a standard metadata JSON. There is no executable code, no instructions for data exfiltration, unauthorized network calls, persistence mechanisms, or any other malicious activities. The `boundaries.md` file even explicitly defines restrictions on external communication and actions requiring user permission, indicating a focus on controlled behavior. The content is fully aligned with its stated purpose of defining an agent's identity.
Findings (0)
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.
An agent using this identity guidance may proceed with in-scope workspace work without asking each time.
The skill defines cases where an agent may act without asking, including technical work and background maintenance inside a workspace. This is relevant to users because it could shape tool-using agents to make workspace changes autonomously.
### Act Without Asking - Technical implementation within approved scope - Research, documentation, internal work - Fixing problems the agent created - Advancing already-greenlit tasks - Background maintenance **The line:** If it only affects workspace → act.
Define exactly what counts as approved scope, require approval for destructive or bulk changes, and ask the agent to summarize autonomous actions after completion.
The agent may adapt based on inferred preferences and prior interactions, which can be helpful but may also preserve incorrect or sensitive assumptions.
The skill encourages maintaining a user model and remembering preferences, boundaries, and relationship history. This is aligned with the adaptation purpose, but persistent personalization can influence future responses and should be user-controllable.
When corrected, ask internally: - Knowledge gap? → Learn the fact - Judgment error? → Calibrate confidence - Unknown preference? → Update user model ... ### When to Remember - Core communication preferences - Proven effective approaches - Established boundaries - Relationship history and trust level
Keep learned preferences visible, allow users to correct or reset them, and avoid storing sensitive inferences unless explicitly approved.