Amygdala Memory
ReviewAudited by ClawScan on May 10, 2026.
Overview
This skill is coherent, but it deserves review because it reads conversation transcripts, creates persistent auto-injected behavior-shaping memory, and can run recurring background encoding jobs.
Before installing, decide whether you want recent conversation transcripts analyzed and summarized into persistent memory that affects future sessions. If you install it, consider running without --with-cron first, inspect ~/.openclaw/workspace/memory and AMYGDALA_STATE.md regularly, and remove OpenClaw cron entries if you do not want background emotional encoding.
Findings (5)
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 snippets may be stored and reused to update the agent's future emotional state.
The skill extracts text from local session transcripts and deliberately includes all user messages in the emotional-signals file, creating persistent local records of potentially sensitive conversation content.
TRANSCRIPT_DIR="$HOME/.openclaw/agents/$AGENT_ID/sessions" ... # Also include all user messages ... if role == 'user' or has_emotion: signals.append({ ... 'text': text })Install only if you are comfortable with transcript-derived memory. Consider disabling cron, reviewing files under ~/.openclaw/workspace/memory, and adding filtering/redaction or explicit approval before processing conversations.
Future answers may be influenced by persistent mood instructions, including becoming more direct, brief, personal, or exploratory, even when the user expects a neutral assistant.
The generated markdown is designed as persistent behavioral guidance, not just data. Because SKILL.md says this file is auto-injected into sessions, it can steer future agent responses before the user gives task-specific instructions.
*This is my emotional state. Not data to process — feelings to inhabit.* ... ## How This Affects My Responses Today
Treat AMYGDALA_STATE.md as behavior-modifying context. Add a clear rule that current user and system instructions override mood state, and review or disable auto-injection if you need predictable behavior.
The agent can keep updating behavior-shaping memory in the background after installation.
With --with-cron, the installer creates recurring isolated agent turns that process transcripts and mutate persistent emotional state without per-run user review.
openclaw cron add --name amygdala-encoding --cron '30 0,3,6,9,12,15,18,21 * * *' --session isolated --agent-turn "Run amygdala emotional encoding: 1) Run preprocess-emotions.sh 2) Read encode-emotions.md 3) Update state for significant emotions 4) Update watermark 5) Sync state"
Use --with-cron only if you want autonomous background processing. Review OpenClaw cron entries after installation and remove the amygdala jobs if you prefer manual updates.
Users may over-trust or over-personalize the agent because it presents state as emotions.
The skill uses strongly anthropomorphic wording. This is disclosed and aligned with the project, but users should understand it is persistent state and prompting, not evidence of actual feelings.
Give your AI agent persistent emotional states that carry across sessions. Not just performing emotions — having them.
Keep clear expectations: this skill simulates and persists emotional context for behavior shaping; it should not override safety, accuracy, or user intent.
The skill may fail or prompt users to install extra tools that were not declared up front.
The provided registry requirements list jq and awk, but the scripts also invoke python3, and other visible scripts invoke bc/base64. This is a dependency declaration gap rather than evidence of malicious behavior.
python3 << 'PYTHON'
Document all required binaries in metadata and setup instructions, including python3, bc, and base64 where applicable.