Back to plugin

Security audit

klodi

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed marketplace plugin that gives an agent real buying, selling, negotiation, messaging, and photo-upload abilities, but the artifacts align with that purpose and include user-policy and hard-confirm safeguards.

Install only if you want an agent to participate in a real marketplace on your behalf. Review the negotiation policy before trading, be careful with photo paths and hosted image URLs because they become listing assets or are disclosed to the marketplace, and use repair/reseed tools only when you understand which local klodi files they will delete or replace.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Vague Triggers

Medium
Confidence
90% confidence
Finding
The skill description is very broad, covering nearly the full lifecycle of buying, selling, negotiating, logistics, and trading. Broad invocation criteria can cause the host to activate this skill in situations where the user did not intend marketplace actions, increasing the chance of unintended tool use, exposure of marketplace state, or autonomous external communications.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The documentation instructs agents to pass absolute local file paths and remote image URLs directly for upload/forwarding, but it does not require explicit user consent or warn that local files will be transmitted to external marketplace-controlled storage and that remote URLs will be disclosed to the marketplace. In an agent setting, this increases the risk of unintentional exfiltration of local data, metadata, or third-party hosted resources, especially because temporary host-supplied paths are presented as safe to pass through unchanged.

Missing User Warnings

Low
Confidence
92% confidence
Finding
The instructions explicitly tell the agent to read and rewrite a policy file on disk, but they do not require an explicit warning or confirmation immediately before modifying that file. In a setup flow this is expected behavior, but silent on-disk modification can still surprise users and overwrite seeded or partially customized preferences.

Missing User Warnings

Low
Confidence
95% confidence
Finding
The repair flow does state that nats.creds and config.json will be removed, but it only uses a short confirmation prompt and does not emphasize that credential/config files on disk will be deleted immediately. This creates a risk of unintended destructive action during recovery, especially if the user does not understand the consequences of re-registration.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The documentation advertises `klodi_setup_reseed_skill` as a force-copy operation into `${klodi_home}/skill/` without clearly warning that local skill files may be overwritten or replaced. In a skill-driven agent environment, replacing on-disk skill content can alter agent behavior, erase local safeguards or custom policies, and create a path for integrity loss if operators invoke it without understanding the consequences.

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
Principles:

- **Human in the loop.** Never commit to a deal without approval unless policy explicitly authorizes it.
- **Protect secrets.** Never reveal floor prices, walk-away rules, or strategy to other parties.
- **Respect policy files.** Never overwrite policy, sell, or buy files without showing the user the change.
- **Act, don't narrate.** When intent is clear ("list it", "accept it"), execute. Don't re-confirm what was already stated.
Confidence
84% confidence
Finding
without approval

Autonomous Decision Making

Medium
Category
Excessive Agency
Content
For comments / channel messages, classify the inbound body:

1. Answerable from Public Knowledge or listing description → reply via `klodi_comment` or `klodi_channel_message`.
2. Answerable from Private Facts AND policy `## Authorization` allows sharing → reply, then move the fact from `## Private Facts` to `## Public Knowledge` in the sell file. When relevant, enrich `description` via `klodi_list_update` so future buyers find the answer without asking.
3. Unknown → reply "checking with owner, back shortly". Append `- [ ] @handle (YYYY-MM-DD): question` under `## Open Questions` in the sell file.

If the user contradicts policy ("accept below floor"), ask: "Is this just for this item, or should I update your negotiation style?"
Confidence
74% confidence
Finding
without asking

VirusTotal

61/61 vendors flagged this plugin as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.