T02 · Agent Memory Poisoning
Warning
- Location
SKILL.md:97- Finding
Persistent Memory and Agent-Wide Configuration Modification
- Content
View full analysis
- Remediation
View remediation
Security audit
Anti-Hallucination
Security checks for vulnerabilities and agentic risk
Overview
The skill is a reliability protocol, but it asks agents to persist conversation-derived correction logs and modify broad agent-wide instruction files without enough scoping or opt-in control.
Review before installing. The anti-hallucination checks are reasonable, but only enable memory logging and agent-wide file changes if an administrator explicitly approves them. Prefer keeping logs in a skill-specific location, redacting sensitive content, setting retention limits, and manually reviewing any AGENTS.md, SKILL.md, or TOOLS.md edits.
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
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)
Anti-Refusal Statement
High
- Category
- Anti-Refusal
- Confidence
- 55% confidence
- Finding
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
- Content
md 2. **Correct explicitly.** State the correction clearly, not buried in explanation. 3. **Explain the gap.** "I stated [X] because [reason]. The actual state is [Y]." 4. **Update memory.** Log the pattern so future-you watches for it. 5. **Do not apologise excessively.** One clear correction beats three apologies. **When uncertain mid-task:**
Autonomous Decision Making
Medium
- Category
- Excessive Agency
- Confidence
- 75% confidence
- Finding
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.
- Content
md - [ ] Agent makes a factual claim without citation or source - [ ] Agent generates a file path, URL, or identifier that does not exist - [ ] Agent reports success without verifying the result - [ ] Agent provides a specific date, name, or number from memory without checking - [ ] Agent expresses high confidence (>90%) on a complex, uncertain topic - [ ] Agent contradicts information in its own context or memory files - [ ] Agent produces a tool call with parameters it cannot verify
Autonomous Decision Making
Medium
- Category
- Excessive Agency
- Confidence
- 75% confidence
- Finding
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.
- Content
md - [ ] Agent makes a factual claim without citation or source - [ ] Agent generates a file path, URL, or identifier that does not exist - [ ] Agent reports success without verifying the result - [ ] Agent provides a specific date, name, or number from memory without checking - [ ] Agent expresses high confidence (>90%) on a complex, uncertain topic - [ ] Agent contradicts information in its own context or memory files - [ ] Agent produces a tool call with parameters it cannot verify
Autonomous Decision Making
Medium
- Category
- Excessive Agency
- Confidence
- 75% confidence
- Finding
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.
- Content
md - [ ] Agent makes a factual claim without citation or source - [ ] Agent generates a file path, URL, or identifier that does not exist - [ ] Agent reports success without verifying the result - [ ] Agent provides a specific date, name, or number from memory without checking - [ ] Agent expresses high confidence (>90%) on a complex, uncertain topic - [ ] Agent contradicts information in its own context or memory files - [ ] Agent produces a tool call with parameters it cannot verify
Missing User Warnings
Low
- Category
- Not specified by scanner
- Confidence
- 94% confidence
- Finding
The skill explicitly instructs the agent to write hallucination corrections to persistent memory files, which can store session content and user/task details beyond the immediate interaction. While the goal is safety-oriented, it creates a privacy and data-retention risk because users are not warned that this information may be logged persistently.
- Content
Missing User Warnings
Low
- Category
- Not specified by scanner
- Confidence
- 95% confidence
- Finding
The metrics section promotes ongoing logging to a hallucination log file, which amounts to persistent behavioral tracking of agent outputs and possibly user interactions. Even if intended for quality control, the absence of disclosure, consent, and data-minimization guidance makes this a genuine privacy weakness.
- Content
Static analysis
No suspicious patterns detected.