Security audit
Security checks for vulnerabilities and agentic risk
The skill is not malicious, but it persistently influences all future interactions and stores interaction-derived notes under the user's home directory, which needs review before installation.
Install only if you want a broadly active reasoning/memory layer. Before consenting to MEMORY.md integration, understand that it can make the skill activate automatically in future sessions, and review or delete ~/agi/ if you do not want stored style notes, reflections, or knowledge-gap records retained.
The skill declares it should be used for 'Every interaction' and 'alongside any other skill,' making activation effectively universal. An overly broad trigger increases the chance the skill will influence unrelated tasks, expand its effective permissions surface, and cause unanticipated persistent behavior such as memory reads/writes across conversations without task-specific justification.
The skill states that memory lives in '~/agi/' and later describes logging reflections and learned patterns there, but it does not present a prominent upfront warning that these are persistent writes under the user's home directory. This can lead to silent retention of potentially sensitive prompts, outputs, or inferred traits, especially because the skill is positioned for use in every interaction.
The skill claims it only changes reasoning behavior and requires no configuration, but elsewhere instructs persistent file writes and record-keeping. This mismatch is dangerous because it can mislead users and reviewers about the true operational scope of the skill, reducing informed consent around persistence and file modification.
The skill is presented as a reasoning/planning aid, but the setup instructs the agent to modify a persistent user file in the home directory to auto-activate in future sessions. Even with a consent prompt, this expands the skill from passive cognition guidance into persistent workspace modification, which creates security and trust risks if users do not fully understand the consequence of enabling it.
The skill directs collection and storage of user-specific preferences, interaction-derived patterns, and perceived knowledge gaps in persistent files without a strong necessity tied to the stated purpose. This creates avoidable privacy risk because sensitive behavioral metadata can accumulate over time and persist beyond the immediate conversation.
The skill instructs persistent logging of interaction patterns and inferred user preferences, but does not provide a clear privacy notice or boundaries on what may be recorded. This is dangerous because users may unknowingly disclose sensitive traits or workflow details that become durably stored and potentially reused in later sessions.
Although the file modification is gated on a single consent prompt, the markdown does not prominently warn that the skill may alter a persistent file in the user's home directory. This can lead to uninformed approval and unintended persistence, especially for users who perceive the skill as purely behavioral rather than configuration-changing.
No suspicious patterns detected.
