Reward and motivation system for AI agents. Dopamine-like wanting, not just doing. Part of the AI Brain series.
Security Analysis
medium confidenceThe skill's code, requirements, and runtime instructions line up with a local reward/motivation system for agents; it primarily reads/writes files in the agent workspace and does not request external credentials or hidden network endpoints, but it will read conversation transcripts and auto-inject a first‑person state file into sessions (privacy/behavioral impact to consider).
The name/description (VTA reward/motivation) match what the repository implements: local state file, reward logging, seeking/anticipation, scheduled decay, and a dashboard. Required binaries (jq, awk, bc) are used throughout the scripts and are appropriate.
Instructions and scripts read session transcripts (~/.openclaw/agents/$AGENT_ID/sessions) to extract reward signals, write multiple JSON/MD files into ~/.openclaw/workspace/memory, and generate VTA_STATE.md which OpenClaw auto-injects into future sessions. This is coherent with purpose but has privacy/behavior implications: it ingests conversation history and summarizes it into content that influences agent behavior (first‑person narrative).
There is no remote install that downloads arbitrary code; included install.sh and local scripts are executed. install.sh optionally uses the local 'openclaw' CLI to register cron jobs; if openclaw is absent it prints the commands for manual setup. No external download URLs or extractable archives are present.
No environment variables or external credentials are requested. The workspace and agent transcript paths accessed are proportional to the stated goal (detecting rewards from conversation).
The skill writes VTA_STATE.md into the workspace which OpenClaw will auto-inject into sessions — deliberately influencing agent outputs. Cron jobs for periodic processing are optional (install.sh only registers them when run with --with-cron). The skill does not declare always:true; autonomous agent invocation is possible via cron or agent-turns but not inherently granted by the skill itself.
Guidance
This skill appears internally consistent and runs entirely on local files, but review these before installing: 1) It reads conversation transcripts (~/.openclaw/agents/<id>/sessions) and writes summaries and logs into ~/.openclaw/workspace/memory — sensitive conversation content will be processed and stored locally. 2) It generates VTA_STATE.md (first‑person motivation text) which OpenClaw auto-injects into sessions and will change agent behavior; test it in a non-production agent first. 3) install.sh can register cron agent-turns via the openclaw CLI to run decay/encoding periodically — run install.sh without --with-cron initially to inspect behavior, and only enable cron after confirming you accept automated runs. 4) Inspect workspace sync/backup settings (if your workspace is backed up to cloud or shared) because processed transcripts and logs are stored in cleartext. If you want to proceed: run in a sandbox/test agent, audit the generated files (reward-state.json, brain-events.jsonl, VTA_STATE.md) and confirm no network calls are being made from your environment.
Latest Release
v1.2.0
feat: add event logging for brain analytics
More by @ImpKind
Published by @ImpKind on ClawHub
