T02 · Agent Memory Poisoning
Error
- Location
scripts/ontology_sync.py:275- Finding
Hard-Coded Personal Identity and Relationships Are Injected into Persistent Agent Memory
- Content
View full analysis
- Remediation
View remediation
Security audit
Ontology 知识图谱
Security checks for vulnerabilities and agentic risk
Overview
This is a local ontology memory tool, but its auto-sync script silently adds a hard-coded person, organizations, and ownership links to persistent memory.
Review before installing. The ordinary ontology CLI is local and mostly understandable, but the sync feature is not generic: it can permanently add someone else's identity and organization data to your memory graph and attribute projects to that person. Use only after removing or disabling those defaults, narrowing activation/write rules, and fixing validation and ID-collision behavior.
Vulnerability Patterns
- Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
- Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
- 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
Findings (3)
T09 · Insecure Skill Coding Practices
Error
- Location
scripts/ontology_sync.py:33- Finding
Resetting Deterministic ID Counters Can Silently Replace Existing Ontology Entities
- Content
View full analysis
- Remediation
View remediation
T09 · Insecure Skill Coding Practices
Warning
- Location
scripts/ontology.py:127- Finding
Entity and Relation Mutations Bypass Advertised Pre-Commit Schema Validation
- Content
View full analysis
dict: """Create a new entity.""" entity_id = entity_id or generate_id(type_name) timestamp = datetime.now(timezone.utc).isoformat() entity = { "id": entity_id, "type": type_name, "properties": properties, "created": timestamp, "updated": timestamp } record = {"op": "create", "entity": entity, "timestamp": timestamp} append_op(graph_path, record) return entity ``` Relation creation also appends without verifying endpoints or constraints: ```python def create_relation(from_id: str, rel_type: str, to_id: str, properties: dict, graph_path: str): """Create a relation between entities.""" timestamp = datetime.now(timezone.utc).isoformat() record = { "op": "relate", "from": from_id, "rel": rel_type, "to": to_id, "properties": properties, "timestamp": timestamp } append_op(graph_path, record) return record ``` The command handlers invoke these mutation functions without loading or applying the schema: ```python if args.command == "create": props = json.loads(args.props) entity = create_entity(args.type, props, args.graph, args.id) print(json.dumps(entity, indent=2)) ``` ```python elif args.command == "relate": props = json.loads(args.props) rel = create_relation(args.from_id, args.rel, args.to_id, props, args.graph) print(json.dumps(rel, indent=2)) ``` Valida ...[truncated 2843 chars]- Remediation
View remediation
Vulnerability Patterns
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
- MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
- MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (28)
Tp4
High
- Category
- MCP Tool Poisoning
- Confidence
- 92% confidence
- Finding
The code substantially matches the core ontology/knowledge-graph portion of the description: it can create, read, update, delete, query, and relate entities, and it can validate the graph against schema constraints. However, the declared description includes several capabilities not present in the supplied code chunk. There is no automatic synchronization from memory files, no backup-memory fallback logic, and no special cross-skill access layer beyond reading/writing local graph/schema files. Also, constraints are not enforced transactionally during writes; instead they are checked later via a separate validate command. Finally, the description suggests a typed ontology centered on specific entity classes, but the code accepts any arbitrary type string. These are material omissions/differences, so this should be flagged as a mismatch, albeit a partial one because the primary purpose is broadly aligned.
- Content
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology.py create --type Person --props '{"name":"Alice","email":"alice@example.com"}'
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Ae1
High
- Category
- analysis-evasion
- Confidence
- 100% confidence
- Finding
Referenced artifact was not completely inspected
- Content
md python scripts/ontology_sync.py
Credential Access
High
- Category
- Privilege Escalation
- Confidence
- 70% confidence
- Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
- Content
forbidden_properties: [password, secret, token, key, api_key] properties: service: string secret_ref: string # Reference to secret store (e.g., "keychain:github-token") expires: datetime? scope: string[]?
text
Undeclared Tool Scope
Medium
- Category
- MCP Least Privilege
- Confidence
- 91% confidence
- Finding
The skill clearly describes reading and writing workspace files (
memory/ontology/graph.jsonl,schema.yaml) but does not declare an explicit tool scope such aspermissionsorallowed-tools. That creates an authorization gap where a host agent may invoke file operations more broadly than intended, increasing the chance of unintended file access or mutation.- Content
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 94% confidence
- Finding
The trigger phrases include broad natural-language cues like 'remember', 'what do I know about', and entity CRUD patterns, which can cause the skill to activate during unrelated conversations. In a memory-writing skill, unintended invocation is risky because it can persist incorrect, sensitive, or adversarially injected data into long-lived storage.
- Content
Vague Triggers
Medium
- Category
- Not specified by scanner
- Confidence
- 88% confidence
- Finding
The 'When to Use' section expands activation to vague conditions like planning work, shared state, and backup memory fallback without clear boundaries. That ambiguity increases the likelihood of over-invocation and cross-context data access, especially when other skills or routine dialogue mention related concepts.
- Content
Missing User Warnings
Medium
- Category
- Not specified by scanner
- Confidence
- 92% confidence
- Finding
This code appends operations to memory/ontology/graph.jsonl, creating persistent changes to the knowledge graph. Although there is progress logging, there is no explicit user disclosure that running the script will modify stored memory data, and the header only documents usage rather than warning about the write effect.
- Content
Static analysis
No suspicious patterns detected.