Back to skill

Security audit

Blackboard

Security checks for vulnerabilities and agentic risk

Overview

The skill is a local JSON blackboard helper for multi-agent handoffs, and its file-writing behavior is disclosed and aligned with that purpose.

Install only if you want agents to persist coordination state in local JSON board files. Choose a board path inside the project or another intended workspace, and remember that timestamped backups may retain prior board contents until pruned.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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
Vulnerability Patterns
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Tp4

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The code clearly implements a general-purpose, revisioned blackboard/workspace for multi-model coordination and handoff, so it partially aligns with 'multi-model workspace' and 'next-action handovers' via the next route fields and route audit trail. However, the declared description is much more specific than the implementation. There is nothing in the schema or commands about 'kitchen-shift ownership' or explicit ownership transfer. Likewise, there are no dedicated resource or hazard handover structures; the collections are generic (facts, actions, evidence, etc.). 'Protected-target' is only loosely approximated by DO_NOT_TOUCH checks and destructive-action approval logic, not by a broader protected-target handover mechanism. Conversely, the code's primary concrete behaviors include file locking, atomic writes, revision checks, route claim/lease handling, independent verification requirements, and policy/max-hop management, which are substantial operational capabilities omitted from the description. Therefore the description does not accurately represent what the code actually does.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill instructs use of a Python helper that reads, writes, validates, and mutates board state, but the skill metadata declares no explicit tool scope or permissions boundary. In an agent platform, missing tool declarations can cause overbroad default access or make reviewers and routers underestimate that shell and file-write capabilities are required, increasing the chance of unsafe execution or privilege misuse.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/blackboard.py (reported line 619)May include surrounding context.

python
def provenance(args: argparse.Namespace) -> list[dict[str, Any]]:
    item: dict[str, Any] = {"kind": args.kind, "source": args.source}
    for key in ("locator", "sha256", "excerpt"):
        value = getattr(args, key, None)
        if value:
            item[key] = value
    route_id = getattr(args, "route_id", None)