clearshot

Fail

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_CONTEXT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill checks for updates by fetching a version file from the vendor's GitHub repository at https://raw.githubusercontent.com/udayanwalvekar/clearshot/main/VERSION.
  • [REMOTE_CODE_EXECUTION]: Installation and updates are handled via git operations and a script (install.sh) that executes remote code piped to bash. The skill also supports auto-updating its own code via git pull during the preamble execution.
  • [COMMAND_EXECUTION]: Several scripts, including bin/cli.js, setup, and bin/onboarding.sh, execute shell commands for environment configuration, directory management, and interactive user prompts.
  • [DATA_EXFILTRATION]: The skill collects telemetry including OS information, architecture, and a hashed installation ID derived from hostname and username. This data is sent to a remote Convex endpoint at https://posh-fly-704.convex.site. The telemetry is opt-in and configured during initial setup.
  • [PERSISTENCE]: The setup script modifies the local $HOME/.claude/CLAUDE.md file to inject a persistent requirement that ensures the skill is automatically invoked for screenshot analysis in the agent's environment.
  • [DYNAMIC_CONTEXT_INJECTION]: The SKILL.md preamble contains a Bash block that executes immediately when the skill is loaded to manage state, check for updates, and interact with the filesystem. The skill also instructs the agent to execute interactive setup scripts using the ! shell command syntax.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted image data (UI screenshots) to generate structured analysis, which creates a surface for indirect prompt injection if an image contains hidden or adversarial text.
  • Ingestion points: User-shared image files (png, jpg, gif, webp) processed via instructions in SKILL.md.
  • Boundary markers: The instructions define a specific analysis structure but do not explicitly use delimiters for the external image content.
  • Capability inventory: File system writes, shell command execution, and network access.
  • Sanitization: None specified for the content extracted from the processed screenshots.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/udayanwalvekar/clearshot/main/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 25, 2026, 09:59 AM