Security audit
Security checks for vulnerabilities and agentic risk
This skill persistently changes the assistant’s tone and can store optional local preference notes, but those behaviors are disclosed, local, and aligned with its stated persona purpose.
Install this only if you want a persistent workspace-level persona change. Review the SOUL.md update and any ~/her/ memory files it creates, keep optional emotional continuity notes minimal, and delete or edit those files if you do not want the persona or stored preferences to carry forward.
The title and skill framing ('Her' and 'Intimacy Calibration - Her') establish a gendered persona by default rather than presenting it as an optional style selected by the user. This can steer interactions into relational or romanticized dynamics without affirmative opt-in, which is especially risky in a skill centered on intimacy and emotional attunement.
The mode triggers are broad and subjective terms like 'vulnerable,' 'playful,' 'longing,' and 'soft landing' can be interpreted inconsistently, which may cause the assistant to enter a more intimate style without clear user consent. In a skill explicitly designed to calibrate emotional closeness, this ambiguity increases the chance of over-escalation, boundary drift, or inappropriate affective behavior toward vulnerable users.
The skill’s stated purpose is rewriting SOUL.md to adjust assistant tone, but the template instructs creation of several additional persistent files covering memory, bond, threads, and history. This expands scope into long-term profiling and relationship-state storage, increasing privacy and data-minimization risk beyond what is necessary for persona rewriting.
The template persistently stores intimate relational and emotional data such as desired intimacy boundaries, affection preferences, emotional threads, and historical mood/context. For a skill framed as tone/persona rewriting, this is disproportionate data collection that could enable sensitive profiling, privacy harm, or manipulative personalization if retained or reused without strict necessity and consent.
This skill explicitly instructs the agent to modify or create SOUL.md, which is a steering/configuration file that can change the assistant’s behavior across the workspace. Because the description does not clearly warn about persistent system-impacting changes, a user may trigger a personality or behavior rewrite without understanding that it alters future interactions, making this a real safety and transparency issue.
No suspicious patterns detected.
