Back to skill

Security audit

genor-orchestrator

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real orchestration dashboard skill, but it exposes powerful local data controls without enough access control or containment.

Install only if you are comfortable with a local orchestration tool that stores project/session history on disk and can run a dashboard service. Keep the dashboard bound to localhost or behind authentication, avoid PM2/cron until you review the exact commands, set ORCHESTRATOR_DATA_DIR deliberately, and do not log secrets or full sensitive prompts unless you are comfortable retaining them locally. VirusTotal was clean; the Review verdict is based on artifact-backed access-control and persistence concerns, not malware telemetry.

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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (12)

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The server binds to 0.0.0.0 and exposes unauthenticated CRUD APIs for models, configuration, and project documents. In a sidecar dashboard context, this is dangerous because any reachable client can read sensitive operational data and modify orchestration state, potentially disrupting routing or planting malicious content for downstream agents.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The project document endpoints allow unauthenticated read, write, and delete operations on files under project directories. In an orchestration skill, those documents likely influence agent context and behavior, so an attacker could tamper with prompts, notes, or state to cause data loss, workflow manipulation, or indirect prompt injection into downstream agent runs.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The onboarding flow instructs the agent to check broad default project directories and enumerate found projects, which can expose unrelated repositories and sensitive local context beyond what is necessary for initial orchestration setup. In an agent skill, this kind of filesystem discovery increases privacy risk and can lead to over-collection of user data if performed before explicit, granular consent.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Installing a cron job introduces persistence and recurring execution on the user's system, which goes beyond a one-time onboarding workflow and can create ongoing network activity or data collection. Even if user approval is mentioned, the instruction lacks safeguards about what will run, how to review it, and how to remove it, making it a real security and trust concern.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Broad conversational triggers like 'start dashboard', 'check prices', or 'onboard project X' can be activated by ordinary user phrasing and cause shell execution or service startup without a clear confirmation boundary. In a skill with filesystem writes, network access, and background process management, accidental trigger collisions materially raise the risk of unintended actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill persistently logs sessions, decisions, project context, and recovery data to disk, but the description does not prominently warn users that their prompts, project metadata, and workflow history may be retained outside ephemeral session storage. Because hooks auto-log and auto-inject context, users may unknowingly expose sensitive codebase or session information to persistent storage and downstream prompts.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The instruction to 'probe every provider you can reach' is overly broad and underspecified, encouraging opportunistic discovery of services, local endpoints, CLIs, and hardware without defined boundaries. In practice, this can lead to unintended access attempts, privacy-invasive enumeration, or interaction with services the user did not intend to expose to the skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The onboarding sequence includes localhost network probing, CLI presence checks, and GPU/system inspection, but does not clearly warn the user that these actions inspect local services and system configuration. Without an upfront privacy and system-impact notice, users may unknowingly authorize broader reconnaissance than expected during setup.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

The instructions direct the skill to write catalog, summary, and session log files without clearly notifying the user that local files will be created or modified. While lower severity than probing or persistence, silent local state changes can still surprise users, overwrite data, or create records containing sensitive model and project metadata.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The cron installation step changes system behavior persistently, yet the user-facing guidance does not adequately warn about the ongoing background execution and its consequences. Persistent scheduled execution is especially sensitive in an agent context because it can continue making network requests or modifying local state after onboarding has ended.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This script persistently records full task text, notes, and optional context/context-file contents into session logs and per-project JSON without any filtering, minimization, or warning. In an orchestration skill, those fields can easily contain prompts, credentials, internal URLs, tokens, proprietary code fragments, or other sensitive operational context, so the behavior creates a real data-exposure risk if the log directory is shared, backed up, synced, or later accessed by other tools or users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script automatically executes a secondary script, discover-models.sh, if present, without presenting its contents, requesting confirmation, or constraining its privileges. In a skill package context where auxiliary scripts may change over time, this creates a trust boundary issue: running onboarding can trigger additional code execution with the user's privileges and side effects not disclosed by this helper.

Content

No source excerpt is available for this finding.