ResonanceEngine

Conversational Frequency Matching — reads invisible micro-signals in every conversation and tells the bot exactly how to respond for maximum engagement, conv...

This package appears to implement what it claims: algorithmic analysis of conversation text and generation of tuning instructions. Before installing or deploying: 1) Review and test the code locally (the skill ships source and tests) to verify behavior. 2) Be cautious about using result.recommendation.to_prompt_injection() to automatically mutate your system prompt — that is effectively a prompt-injection primitive and can change model behavior or bypass safety rules; instead, manually review or sandbox injected fragments and ensure your safety/policy prompts remain authoritative. 3) If you plan to pip install from PyPI or clone the GitHub repo, verify the package name and repository (the SKILL.md example repo URL looks possibly malformed). 4) Consider legal/ethical implications of deploying automated persuasion/monetization logic (consent, transparency, regulated domains). 5) If you need stronger assurance, run the included tests and consider a security review or running the code in a restricted environment before production use.

Type: OpenClaw Skill Name: openpaw Version: 0.1.0 The skill is classified as suspicious due to its core design feature: the explicit use of 'prompt injection' to modify the bot's system prompt. The `SKILL.md` instructions and the `TuningRecommendation.to_prompt_injection()` method in `openpaw/models/metrics.py` demonstrate that the skill generates dynamic text intended to be appended to the agent's system prompt. While the generated content (e.g., conversational advice like 'User is at peak resonance — present the offer/solution now.') is not overtly malicious, this design pattern introduces a significant prompt injection vulnerability. A malicious user could potentially craft input that, when analyzed by the skill, causes the generated 'tuning recommendation' to contain harmful instructions for the LLM agent, leading to unintended actions or information disclosure. This represents a risky capability without clear malicious intent within the provided code.