ClawPsyche
ReviewAudited by ClawScan on May 10, 2026.
Overview
ClawPsyche appears to be a local affect/memory simulator that intentionally changes the agent’s tone and persists per-user state, with no artifact evidence of credential use, network exfiltration, or destructive behavior.
Install this only if you want the agent to maintain a persistent mood and relationship model that can influence future replies. Keep the state file local, avoid real personal identifiers, periodically reset or inspect the JSON state, and ensure the generated snapshot never overrides core safety or task instructions.
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.
The agent’s tone and behavior may be influenced by generated mood flags rather than only by the current user request.
The skill intentionally makes generated affect state part of the agent’s ongoing instruction context every turn. This matches the stated behavior-modulation purpose and includes a prompt-injection caution, but it can still shape future responses.
Run this skill **every turn** ... Inject into your system prompt (Safely) ... wrap it in strict XML tags at the very end of the system prompt
Enable only if you want persistent affect modulation; keep the snapshot separated from core safety and user instructions, and do not let it override higher-priority policies.
Emotional labels and relationship state can carry into later turns and may affect how the agent responds to the same user.
The skill stores persistent local state tied to a user identifier and uses it for relationship/emotional tracking across interactions.
--state ~/clawd/soul/psyche_state.json ... --user "user_id" ... User identifier for relationship tracking
Use non-sensitive user IDs, keep the JSON state file local, review or delete it when needed, and provide a clear reset workflow.
Users have less provenance information and may need to verify the bundled scripts and local runtime manually.
The registry metadata provides little provenance and does not declare the Python runtime even though the skill instructions run bundled Python code.
Source: unknown; Homepage: none; No install spec — this is an instruction-only skill; Required binaries ... none
Install only from a trusted source, inspect the bundled code before use, and declare the Python runtime/install expectations in metadata.