Security audit
Security checks for vulnerabilities and agentic risk
This is a local memory-setup skill that persistently stores user and project notes as advertised, with privacy considerations but no hidden execution, exfiltration, or destructive behavior found.
Install this only if you want the agent to keep local long-term memory files. Review MEMORY.md, memory/feedback.md, and daily notes periodically, and avoid saving sensitive personal, regulated, or confidential information unless you intentionally want it retained.
The skill clearly instructs persistent writes to local memory files (MEMORY.md, memory/feedback.md, daily notes) but does not declare any tool scope or permissions boundary. That mismatch weakens policy enforcement and transparency, making it easier for an agent runtime to perform file writes without explicit review or least-privilege constraints.
The skill directs the agent to store personal information, preferences, and relationships in long-term memory, but provides no user-facing consent notice or privacy warning. This creates a privacy risk because sensitive personal data may be retained locally without the user's clear awareness or approval.
Persisting personal information and relationship details in long-term memory without clear minimization or consent boundaries can lead to unnecessary collection of sensitive user data. Because the skill frames such information as normal memory content, it increases the chance of over-collection and long-lived retention beyond what is needed for task completion.
The write rules instruct the agent to immediately persist human corrections and confirmations, but omit a warning that conversation content may be stored in local files. Users may reasonably assume conversational guidance is ephemeral, so silent persistence increases the risk of retaining sensitive or regulated information without informed consent.
The instruction to record corrections and confirmations 'NOW' encourages default, immediate persistence of user interaction content without a filtering step for sensitivity or necessity. This can result in broad retention of conversation-derived data, including preferences, mistakes, or sensitive context that should not be stored long-term.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
Memories are **long-term assets**, not consumables. They don't expire. But some memories need verification before use:
- **Personal preferences, relationships, history events** — Generally stable. No verification needed.
- **Technical state (service configs, bug status, installed versions)** — Verify before acting on them. The world changes.
- **Before recommending based on memory** — If the memory names a specific file, check it exists. If it names a function or flag, grep for it. "The memory says X exists" is not the same as "X exists now."
The heartbeat routine establishes an ongoing mechanism to capture 'significant events from this conversation' into daily notes, normalizing continuing retention of conversational data. Without tight minimization and consent boundaries, this creates a durable log of user interactions that may accumulate sensitive details over time.
No suspicious patterns detected.