Vta Memory
ReviewAudited by ClawScan on May 10, 2026.
Overview
Review recommended: this is a coherent motivation-memory skill, but it can automatically scan conversation transcripts and inject persistent behavior-changing state into future sessions.
Install only if you want an agent-wide persistent motivation system. Avoid --with-cron until you are comfortable with recurring transcript processing, and review the files under ~/.openclaw/workspace/memory plus VTA_STATE.md to see what conversation-derived data is being stored and reused.
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.
Private conversation content can be copied into persistent workspace memory and reused by the skill, even when the user did not explicitly select those messages as rewards.
The script reads the latest local session transcript and writes conversation text, including all user messages after the watermark, into reward-signals.jsonl for later memory processing.
TRANSCRIPT_DIR="$HOME/.openclaw/agents/$AGENT_ID/sessions" ... if role == 'user' or has_reward: signals.append({ ... 'text': text })Make transcript processing explicitly opt-in, limit it to clearly selected/current-session content, add redaction and retention controls, and show users what will be stored before syncing memory.
If misleading, sensitive, or instruction-like text enters the reward state, it can persist and influence later agent behavior across sessions.
Conversation-derived reward, seeking, and anticipation state is rendered into a markdown file that is meant to be injected into future sessions as behavioral context.
*This is my drive/reward state. Not data — lived motivation.* ... ## How This Affects My Behavior
Treat generated memory as untrusted advisory data, sanitize or summarize user-derived text, avoid instruction-like phrasing, and provide a simple way to review, edit, or delete VTA_STATE.md.
The skill can keep running periodically in the background and continue modifying memory and future session context unless the user removes the cron job.
When installed with --with-cron, the skill creates a recurring autonomous agent job that preprocesses transcripts, detects rewards, logs them, and updates state.
openclaw cron add --name vta-encoding --cron '45 0,3,6,9,12,15,18,21 * * *' --session isolated --agent-turn "Run VTA reward encoding: 1) Run preprocess-rewards.sh ..."
Enable cron only if you want ongoing background processing; the skill should document exact cron effects, provide disable/uninstall commands, and consider requiring review before each encoding run.
Future responses may become more proactive or less ambitious based on stored drive state, which may not always match the user's immediate request.
The generated state file intentionally tells the agent to adjust its behavior based on motivation level.
Proactive mode — suggest improvements, spot opportunities ... Keep tasks simple until I get a win
Ensure user instructions always take priority over motivation state, and review or remove VTA_STATE.md if the agent starts behaving in unwanted ways.