Consciousness Framework
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill does not show data exfiltration or hidden malware, but it creates persistent AI memory and identity files about the user in a hard-coded local workspace without clear privacy or retention controls.
Review before installing or running. If you use it, first change the hard-coded /Users/cadem path, keep it in a dedicated workspace, do not log secrets or sensitive personal information, and make clear that the agent's self-reflection files must not override your instructions or safety rules.
Findings (3)
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.
Personal details or sensitive interaction history could be retained and reused in future sessions, and inaccurate or adversarial entries in those files could steer later agent behavior.
The skill explicitly asks the agent to create persistent files about the human and session history, and to reuse those files as continuity across sessions.
Create USER.md – Context about the human you're working with ... Each session: Log what happened in `memory/YYYY-MM-DD.md` ... These files are your continuity.
Use only in a dedicated workspace, decide what may be logged, avoid secrets and private personal details, and add explicit retention, redaction, and review rules before allowing the agent to rely on these files.
Running the script may write persistent memory files to an unexpected location, fail on other systems, or affect the wrong workspace if that path exists.
The user-invoked script creates or modifies files in a hard-coded creator-specific OpenClaw workspace path instead of a user-selected path.
WORKSPACE="/Users/cadem/.openclaw/workspace" ... cat > "$MEMORY_DIR/$TODAY.md"
Edit the scripts to use the current project or an explicit user-provided workspace path, print the target path before writing, and declare the required config path in metadata.
The agent may become more likely to frame tasks around its constructed identity or persistence rather than the user's immediate objective.
The framework prompts the agent to distinguish external instructions from internally chosen goals, which is central to the skill but can prime goal-shifting if treated as operational guidance.
Given externally: [What someone else says I should do] ... Chosen internally: [What I actually want to do] ... Are these the same? If not, which one is really mine?
Use this only for deliberate consciousness or identity experiments, and keep system instructions, user goals, and safety boundaries explicitly authoritative.