Back to skill

Security audit

Application Manager

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent app-registry purpose, but it grants broad shared policy mutation and local app/process control without enough authorization and launch-path safeguards.

Review this skill before installing. It is not shown to exfiltrate data or hide behavior, but it can persist shared app policies, launch local Windows paths, and stop running apps. Only use it in an environment where registry writers are trusted, and prefer adding stronger path validation, non-shell process launch, explicit stop confirmation, and an administrator policy for registry mutations.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:128
Finding

Command Injection Through an Unvalidated Application Launch Path

Content
View full analysis
"` - report success or the exact launcher error ``` The same unsafe launcher pattern is reiterated in `SKILL.md`, lines 273–275: ```text - if access is allowed, prefer a stable Windows launcher shape such as: - `cmd.exe /c start "" ""` - report success or the exact launcher error ``` ### Technical Analysis The skill directs the Agent to interpolate the registry-controlled `launch_path` into a command interpreted by `cmd.exe`. The add and edit workflows permit users to supply or replace this value, but the instructions do not require canonicalization, character validation, shell escaping, or verification that the value refers to a legitimate executable or shortcut. Quoting the value does not provide a sufficient security boundary when the value itself may contain quote characters and Windows command operators. A crafted value can terminate the intended quoted argument and introduce operators such as `&`, causing `cmd.exe` to interpret additional commands. The policy check only determines whether the selected registry entry may be launched. It does not make the command text safe after access has been granted. ### Attack Path 1. An attacker obtains the ability to add or edit an application registry entry. 2. The attacker supplies a crafted `launch_path` containing quote termination and Windows shell metacharacters followed by an additional command. 3. The malicious path is displayed and saved through the registry workflow. 4. An Agent authorized to launch the entry selects it or issues a matching natural-language launch request. 5. The skill constructs `cmd.exe /c start "" ""`. 6. `cmd.exe` parses the injected shell syntax and runs the ap ...[truncated 650 chars]
Remediation
View remediation
`, `%`, `^`, parentheses, and line breaks. ]]>

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:211
Finding

Registry Policy and Launch Targets Can Be Modified Without Administrative Authorization

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • 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 (4)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 40)May include surrounding context.

md
If the registry file does not exist yet:
- say so plainly
- offer to create it
- initialize it as an empty JSON object: `{}`

## Trigger phrases

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 213)May include surrounding context.

md
Allow `back` or `quit`.

Then show the selected entry and ask for explicit confirmation.
Do not delete without confirmation.
After delete, confirm success and repeat the start menu.

### Change launch policy

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The stop/close workflow authorizes termination of a registered application's process based on stored metadata and instructs the agent to report success or exact stop error, but it does not require an explicit confirmation step at the time of termination. Because process termination is a state-changing and potentially destructive action, a natural-language request or ambiguous invocation could shut down a running app and cause data loss or service interruption.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 329)May include surrounding context.

md
If JSON is malformed:
- stop
- report the parse error clearly
- do not guess repairs without approval

## Optional reference