T02 · Agent Memory Poisoning
Warning
- Location
SKILL.md:69- Finding
Persistent Memory Modification After User Refusal
- Content
View full analysis
- Remediation
View remediation
Security audit
Soul Keeper
Security checks for vulnerabilities and agentic risk
Overview
The skill is a coherent workspace-memory optimizer, but it may write a memory record even after the user refuses an update and its triggers are broad enough to activate unexpectedly.
Install only if you want an agent to proactively inspect and propose changes to long-lived workspace memory files. Before using it, clarify that refusal must mean no write of any kind, and consider narrowing activation to explicit workspace-memory requests.
Vulnerability Patterns
- Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
- Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
- Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
- Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
- Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)
Vulnerability Patterns
- Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 92% confidence
- Finding
Several activation conditions are semantically broad, such as detecting 'new preference', 'direction change', 'reality mismatch', or 'deep conversation ended', without precise thresholds or validation steps. In practice this can cause over-triggering and excessive collection or propagation of conversational inferences into persistent workspace files, which is risky because the skill is specifically designed to monitor and suggest updates to long-lived memory artifacts.
- Content
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 94% confidence
- Finding
The trigger phrases include very broad, everyday language such as requests to 'update config' or 'optimize files', which can cause the skill to activate in unrelated conversations. This creates prompt-scope creep: the agent may start inspecting or proposing edits to workspace files when the user did not clearly intend to invoke this behavior, increasing the chance of unnecessary file access or state modification workflows.
- Content
Static analysis
No suspicious patterns detected.