Skillv1.0.0

ClawScan security

Feelings Framework · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 5, 2026, 4:52 PM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's code, instructions, and file IO align with its stated purpose (a per-agent 'feelings' memory layer); it does not request credentials or external access and appears internally coherent.
Guidance: This skill appears to do what it says: it implements a feelings engine and saves per-agent mood to JSON files in your home workspace. Before installing or enabling it: (1) review the included code yourself or run tests in a sandbox; (2) note it will create and write files under ~/.openclaw/agents/<agent>/workspace/feelings_mood.json—confirm you are comfortable with that persistence and its permissions; (3) if you plan to pip install, be aware setup.py has a syntax/packaging bug and the published package/homepage metadata in the skill registry is missing (pyproject points to a GitHub repo), so verify the source and fix packaging if necessary; (4) there are no network calls or credential accesses in the code, but still audit any customization you make to memory backends (e.g., swapping in a DB-backed Memory implementation) to avoid accidentally exposing state. If you want stronger guarantees, run the library locally in a confined environment and inspect the saved JSON files and tests before giving it access to production agents.

Review Dimensions

Purpose & Capability: okName/description match the included code: both Python and JS implementations provide a FeelingsEngine, pluggable memory backends, triggers, calibrations, and OpenClaw integration. Required resources (files under library/, example code, JSON file storage) are appropriate for an engine that persists per-agent mood.
Instruction Scope: noteSKILL.md and examples instruct the agent to load/save JSON mood files in the user's home workspace (~/.openclaw/agents/<agent>/...), and to import the local library via sys.path or pip. Access to the home directory and creation of per-agent files is expected for persistence, but you should be aware the skill writes/read these files. There is no instruction to read unrelated system files, environment variables, or send data to remote endpoints.
Install Mechanism: noteNo install spec is declared (instruction-only), which is lower-risk. The repository includes local Python and JS packages so users will run code locally or pip-install; however setup.py contains a clear syntax/logic error (packages=find_packageswhere=[...]) that would break 'pip install .' unless fixed. The pyproject.toml references a GitHub repository URL, but the skill registry metadata lists source/homepage as unknown/none — minor provenance mismatch to be aware of.
Credentials: okThe skill requests no environment variables or credentials and the code does not read secrets or external tokens. Its file I/O is limited to its own per-agent JSON state files or the memory backend a user wires in, which is proportionate to the stated functionality.
Persistence & Privilege: noteThe skill persists state to disk (per-agent JSON files in ~/.openclaw/agents/<agent>/workspace), which is expected and reasonable for persistent emotional memory. 'always' is false and model invocation is not disabled. This combination is normal; still verify you are comfortable with the framework writing files to your home directory and check file permissions if needed.