The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically imports and executes JavaScript walkthrough scripts (.mjs) generated at runtime. In scripts/record.mjs, the use of await import() on a file path provided via command-line arguments allows for arbitrary code execution.
[COMMAND_EXECUTION]: The script scripts/narrate.sh uses eval to process the CLONEY_ARGS environment variable (eval "extra=(${CLONEY_ARGS:-})"). This creates a shell injection vulnerability where a specially crafted string in the environment variable could be used to execute arbitrary shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes walkthrough instructions (.mjs) and narration text (.md) authored by the agent based on user-provided descriptions. This represents an attack surface where instructions embedded in a web application being recorded or within a user's prompt could influence the generated scripts or narration logic.
[EXTERNAL_DOWNLOADS]: The Makefile orchestrates the installation of the Playwright framework and the Chromium browser from official registries. While the sources (Microsoft/NPM) are well-known, the skill requires full network access to download and install these functional binaries at runtime.
[COMMAND_EXECUTION]: The SKILL.md file contains explicit instructions for the agent to execute recording and narration tasks "unsandboxed" to bypass platform security constraints that would otherwise prevent Chromium or hardware acceleration from functioning correctly.

demo-recorder