T02 · Agent Memory Poisoning
- Location
SKILL.md:18- Finding
Persistent Agent Steering Creates a Memory-Poisoning Surface
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is not clearly malicious, but it persistently changes agent memory and workspace steering while relying on missing setup and security documents, so it needs review before install.
Install only if you are comfortable with the agent storing corrections and preferences in local files and letting those records influence future sessions. Before use, require explicit approval for creating memory, promoting rules to always-loaded memory, and modifying workspace steering files; also verify the missing setup and boundary documents are supplied by a trusted source.
SKILL.md:18Persistent Agent Steering Creates a Memory-Poisoning Surface
SKILL.md:33Security-Critical Behavior Depends on Missing Referenced Files
The skill explicitly instructs the agent to automatically log user corrections, preferences, and recurring patterns into local files, but the surrounding user-facing guidance does not present this as a prominent warning or consent checkpoint. This creates a privacy and transparency risk because users may disclose sensitive preferences or behavioral data without realizing it is being persisted across sessions.
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.
- Pattern used 3x in 7 days → promote to HOT
- Pattern unused 30 days → demote to WARM
- Pattern unused 90 days → archive to COLD
- Never delete without asking
### 4. Namespace Isolation
- Project patterns stay in `projects/{name}.md`