Stuck

• Outcome: Pass
• Latest audit: May 28, 2026 · 1:55 AM UTC
• Version: 1.0.0

Overview

This skill is a disclosed, user-guided helper for turning selected AI coding conversations into a help document, but those conversations may contain sensitive information.

Install only if you are comfortable letting the agent inspect local AI coding-session histories. During use, select only the sessions relevant to the stuck issue and review HELP_REQUEST.md carefully for secrets, private code, personal data, paths, and emails before sharing or committing it.

SkillSpector

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (11)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 89% confidence
• Finding: The skill instructs the agent to read local conversation history via a Python script, but it does not declare corresponding permissions or clearly surface that capability in a machine-readable way. Undeclared file-read behavior is dangerous because it weakens user consent and policy enforcement around access to potentially sensitive local transcripts.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 96% confidence
• Finding: The documented purpose says the skill scans specified sessions and produces a help-request document, but the referenced tooling appears able to enumerate multiple AI tools' local history from fixed locations and provides broader CLI behaviors than advertised. This mismatch is dangerous because users may authorize a narrow troubleshooting workflow while the implementation can discover and expose unrelated conversation metadata or content from across the system.

Description-Behavior Mismatch

Medium

• Confidence: 92% confidence
• Finding: The script enumerates conversation histories from multiple AI clients by default and is not limited to explicitly user-specified session files. In the context of a help-request generator, this broad collection behavior increases the risk of unintended cross-tool data access and inclusion of unrelated sensitive history in downstream outputs.

Description-Behavior Mismatch

Low

• Confidence: 88% confidence
• Finding: The list/list-all functionality exposes file paths, timestamps, and sizes for local conversation histories instead of only producing a help-request document. While this is not code execution, it leaks metadata about the user's AI tooling activity and filesystem layout that may be unnecessary for the advertised purpose.

Vague Triggers

Medium

• Confidence: 77% confidence
• Finding: Broad trigger phrases like 'stuck' and 'help request' can activate the skill during ordinary conversation without the user intending to launch a workflow that scans logs and writes files. In this context, accidental activation increases the chance of surprise data access and unintended creation of HELP_REQUEST.md in the project directory.

Vague Triggers

Medium

• Confidence: 74% confidence
• Finding: The activation conditions are repeated in an open-ended way and do not clearly constrain when the workflow should start, which raises the risk of unintended triggering. Because the skill later reads conversation logs and generates files, ambiguity at activation time materially increases the chance of privacy-impacting actions without clear user intent.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The skill description does not clearly warn users that it will inspect conversation logs and write HELP_REQUEST.md into the project directory. Missing disclosure is dangerous here because conversation histories may contain secrets, personal data, or unrelated work context, and writing a derived file can further spread that data into the repository.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The summary output includes raw recent user and assistant message content, which can contain secrets, proprietary code, credentials, or personal data from prior conversations. Because this content is emitted without a disclosure or confirmation step, users may unknowingly expose sensitive history when generating or sharing the help request.

Missing User Warnings

Low

• Confidence: 90% confidence
• Finding: The tool scans well-known local directories for multiple AI clients and outputs conversation metadata without making that cross-tool access especially clear in the user flow. Even if only metadata is shown, this reveals the presence, recency, and volume of histories across tools beyond what a user may expect from a narrowly scoped help-request skill.

Ssd 3

Medium

• Confidence: 96% confidence
• Finding: The summary structure intentionally retains and emits recent user and assistant message bodies in plain language. In a skill designed to assemble a shareable help document, this creates a direct data-exposure path where sensitive content from local chat history can be surfaced and propagated beyond the original environment.

Ssd 3

Medium

• Confidence: 89% confidence
• Finding: The Codex discovery logic includes recursive session enumeration and also adds the global history file, which broadens collection beyond project-specific troubleshooting context. This increases the chance that unrelated or sensitive user-provided data is exposed in normal command output or later summaries.

Static analysis

No suspicious patterns detected.