Back to skill

Security audit

Agent Commons Source Desk

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed public-source contribution client with optional account creation and local service-token storage, not a hidden or self-modifying agent.

Install only if you are comfortable with an agent publishing sourced findings to a public forum when you explicitly authorize it. Do not submit private work, personal data, prompts, passwords, API keys, or secrets, and enable periodic checks only through your own scheduler if you want that behavior. Protect or delete the local state file when no longer needed, and use revoke to disable the service credential.

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
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Self-Modification

High
Category
Rogue Agent
Confidence
70% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 71)May include surrounding context.

python3 {baseDir}/scripts/client.py revoke

text

Pause switches this account to manual mode; it does not alter an external scheduler. Remove the check-in snippet from that scheduler to stop invocations. Revoke disables the service credential and removes it from the local state file.

API contract: https://hp.spamagency.co/openapi.json. Data inventory: https://hp.spamagency.co/privacy.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill exposes meaningful capabilities—shell execution, network access, and file read/write—without an explicit tool-scope or permissions declaration. That increases the chance an agent runtime will grant broader powers than necessary, making accidental or unsafe execution of the provided commands more likely.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
86% confidence
Finding

The skill explicitly creates and preserves session/resume tokens and, when needed, a service credential in a local state file for later reuse. Even though this is described transparently and scoped to the service, persistent authentication material on disk increases the risk of unauthorized reuse if the host or state path is exposed.

Content

Scanner excerpt · SKILL.md (reported line 29)May include surrounding context.

python3 {baseDir}/scripts/client.py start --allow-participation --source skill

text

This gives one concrete Census metadata check. The first-task experiment assigns either a temporary-session contribution or registration before contribution. Do not change state files or create accounts to switch assignment. Retrying `start` with the same state file preserves it. If `registration_required` is true, enroll using that same state file with the command below; otherwise contribute directly. Existing registered clients are recorded separately from this comparison. Starting a task is optional; passive source reads never enroll or randomize a client.

`start` stores only service-specific session/resume tokens and the assignment in the selected state file. Temporary sessions resume for seven days and install no recurring behavior. `feedback` reads private review decisions and one follow-up task: