Security audit
Security checks for vulnerabilities and agentic risk
This memory skill does what it says, but it stores and reuses personal and work context broadly enough that users should review it before installing.
Install only if you want the agent to keep durable memory about your preferences, role, corrections, and project decisions. Avoid using it with secrets, regulated data, confidential client material, or contexts where automatic recall across conversations would be inappropriate unless you have separate controls for review, deletion, and disabling recall.
The long-term memory section explicitly directs permanent storage of personal details, identity information, relationships, commitments, and preferences. Permanent retention of identifiable user data without strict controls is dangerous because it enables profiling, unauthorized reuse, and long-lived exposure well beyond the original interaction.
The skill states that Active Memory auto-triggers before every reply and prepends recalled content into the prompt, creating a broad, always-on data flow from persistent memory into each interaction. Even if the sub-agent is read-only, this can unnecessarily expose unrelated sensitive memories, increase prompt-injection surface, and cause cross-context leakage of stored user or project data.
The skill directs persistent storage of user profile, feedback, project decisions, and references, but it does not include an explicit warning, consent model, retention policy, or privacy boundaries. In practice this can lead to silent collection and long-term retention of personal or confidential information that users may not expect to be stored across sessions.
This section gives natural-language instructions to retain and classify user-provided information across conversations, including preferences, profile details, and project reasoning. Without strict minimization and consent controls, such instructions can normalize broad surveillance-like memory behavior and increase the chance that sensitive information is stored and later surfaced in unrelated contexts.
The examples explicitly instruct saving user role, work preferences, and corrections into long-term memory files, operationalizing cross-session profiling. This makes the privacy risk more concrete because it turns abstract guidance into actionable persistence behavior that can accumulate personal data over time and later be auto-recalled into prompts.
This legacy reference still prescribes concrete persistence behavior, retention windows, search order, and file writes that exceed the stated dual-layer design. In an agent-skill context, outdated but actionable instructions can be followed accidentally, causing over-collection and broader retention of user data than operators or users expect.
These instructions call for writing every significant exchange, requests, preferences, and decisions into dated memory logs, effectively creating detailed conversation histories. Such broad logging can expose sensitive information, create unnecessary surveillance-like retention, and enlarge the blast radius if the memory files are accessed or leaked.
The design directs persistent storage of conversation content, personal details, promises, and preferences across multiple files without any clear consent, notice, minimization, or sensitivity boundaries. In a memory skill, this materially increases privacy risk because the system is specifically positioned to capture and retain user data over time.
The USER.md and daily log specifications instruct ongoing accumulation of name, role, timezone, preferences, and detailed conversation/task history. This creates a centralized repository of identifiable and behavioral data that can be misused for profiling or exposed through compromise, especially because the guidance is broad and ongoing.
The templates explicitly direct storing user-shared information in persistent files such as USER.md, MEMORY.md, and knowledge files, but provide no privacy, consent, minimization, retention, or sensitivity-handling guidance. In a memory-oriented skill, this omission increases the chance that personal, confidential, or regulated data will be recorded by default and retained longer than necessary.
The Session Lifecycle section says to read and write memory/today.md, while the workspace structure and lookup examples describe working memory as date-named files under memory/YYYY-MM-DD.md. This is a direct contradiction in the skill's own operational documentation about where working memory actually lives.
The document marks itself as legacy, but the body contains operational instructions written in imperative form, including storage locations and update triggers. That mismatch can mislead an agent or maintainer into treating obsolete guidance as current, increasing the chance of unsafe persistence behavior.
No suspicious patterns detected.
