Back to skill

Security audit

OverRec Screen

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent for screen capture and window control, but it needs review because it can run repeated screen captures to disk and shows unsafe shell examples for user-supplied window names.

Install only if you are comfortable granting a skill the ability to capture screen regions, save images, draw overlays, and move application windows. For watch mode, require an explicit region, interval, save location, and finite stop condition, and avoid using sensitive screens. Window-title searches should be quoted or otherwise handled as literal arguments, not pasted into shell commands directly.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:97
Finding

Unquoted Window Search Input May Enable Shell Command Injection

Content
View full analysis
OverRec.exe cli window --all # also shows monitor, location, size ``` The unsafe pattern is repeated in the window-placement workflow: ```bash # 1. Find the window OverRec.exe cli window # 2. Snap it (use the WindowID from step 1) OverRec.exe cli snap --windowid ID --location X,Y --size WxH # 3. Optionally draw a brief overlay to confirm placement OverRec.exe cli draw --location X,Y --size WxH --timeout 2 ``` ### Technical Analysis The Skill instructs the Agent to replace `` with a window-title keyword supplied through a user request and then execute the resulting Bash command. The substituted value is not explicitly quoted, validated, or passed through a structured process-execution interface. If the Agent performs direct textual substitution, shell metacharacters in the keyword—including command separators, pipelines, command substitutions, or redirection operators—can be interpreted by Bash rather than passed to `OverRec.exe` as literal window-title text. Quoting only some fixed arguments elsewhere in the document does not protect these unquoted search commands. This weakness is especially relevant because the Skill declares Bash as an allowed tool and explicitly directs the Agent to execute the illustrated commands. ### Attack Path 1. An attacker asks the Agent to find, move, or resize a window. 2. The attacker supplies a crafted “window title” containing shell syntax, such as a title followed by a command separator and an additional local command. 3. The Agent substitutes the supplied text directly for ``. 4. Bash parses the shell syntax instead of treating the complete value as one literal search argument. 5. The injected command executes with the same ...[truncated 984 chars]
Remediation
View remediation
` to state explicitly that direct textual substitution is prohibited. 7. When multiple title words are supported, construct a safely quoted argument array instead of splitting an untrusted string through shell evaluation. 8. Add negative tests covering semicolons, pipelines, command substitutions, redirections, newlines, quotes, and leading hyphens to verify that each value reaches OverRec only as literal data. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger description is broad enough that ordinary requests about the screen, windows, or screenshots could invoke the skill even when the user did not intend to allow screen capture or window manipulation. In this context, over-broad invocation is risky because the skill can access screen contents and move application windows, which can expose sensitive information or disrupt user workflows.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill adds a persistent screen-watch loop that repeatedly captures screenshots and writes them to disk, but this behavior is not reflected in the manifest summary. That hidden expansion of capability materially increases surveillance and data-retention risk because a user invoking a screen-related action may not realize the skill can continuously collect sensitive on-screen data until manually stopped.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The repeated screenshot workflow saves frames to disk in a loop without a prominent warning in the skill description about ongoing capture, storage location, and privacy implications. In a screen-control skill, this is particularly dangerous because screens commonly contain credentials, messages, documents, and other sensitive data that would be silently persisted over time.

Content

No source excerpt is available for this finding.