Security audit
Security checks for vulnerabilities and agentic risk
This skill is purpose-built to remember user preferences, but it does so automatically across every session with silent persistent profiling that users should review carefully before installing.
Install only if you want an always-on preference memory that can write local profile files and influence future conversations automatically. Before using it, review the memory/context-infra files periodically, avoid sharing sensitive preferences you do not want retained, and confirm you are comfortable relying on natural-language retraction rather than a dedicated disable or deletion workflow.
The description states the skill applies learned preferences as defaults in 'every future conversation' without explicit user opt-in. Automatic cross-session behavioral modification is dangerous because it can silently steer outputs, embed stale or sensitive assumptions, and affect unrelated future tasks without the user's informed approval.
The manifest describes the skill as 'Always active in every session,' which is an overly broad activation condition for a capability that reads and writes persistent memory files. This creates continuous, non-consensual profiling behavior across all interactions and expands the attack surface by making the skill run even when unrelated to the user’s task.
The README explicitly says the skill 'silently learns' preferences across sessions and applies them automatically, but it does not present a clear upfront warning or explicit consent mechanism before persistent recording begins. That creates a privacy and transparency risk because users may disclose preference, behavioral, or work-style data without understanding it is being stored in long-lived memory files.
The skill advertises an 'invisible' three-layer memory hierarchy that silently records corrections, preferences, and retractions during normal use. Hidden persistence of user-derived signals is risky because users may not realize they are being profiled, and those records could reveal sensitive habits, opinions, workflows, or decision patterns over time.
The documented design stores raw observations in an append-only log and auto-loads a distilled profile every session, which creates durable accumulation of user-derived data and automatic reuse without review. Append-only storage increases exposure because sensitive or incorrect inferences may persist indefinitely, while auto-loading broadens the chance of unintended influence on future conversations.
The skill instructs the agent to maintain persistent memory, log observations, and distill user preferences into lasting axioms, but it does not clearly warn the user about ongoing file writes or profiling. In context, this is especially risky because the skill is always active and designed to operate silently, enabling covert accumulation of behavioral data without informed consent.
