The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains instructions to modify the Chromium browser binary by replacing it with a wrapper script. This script explicitly adds the --no-sandbox flag, which disables the primary security layer of the browser to allow it to run in restricted environments.
[PRIVILEGE_ESCALATION]: The instructions involve using chmod +x to grant execution permissions to a manually created shell script that wraps the browser binary. It also recommends global installation of packages (npm i -g), which typically requires elevated system privileges.
[CREDENTIALS_UNSAFE]: The skill's workflow requires an APIFY_TOKEN to be passed as an environment variable to record authenticated sessions. While the instructions advise using low-permission test accounts and revoking tokens afterwards, the agent is still tasked with handling and injecting active authentication secrets into the shell environment.
[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing external software, including the @apify/mcpc CLI and @modelcontextprotocol/server-filesystem from the NPM registry. These are vendor-owned or recognized service packages.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to edit and execute VHS .tape files based on user prompts. Because these files act as scripts that drive a real terminal session, there is a risk that untrusted user input could be used to generate malicious commands that are subsequently executed by the vhs tool.
Ingestion points: User requests to create, refresh, restyle, or fix terminal demo tapes in docs/vhs/*.tape (SKILL.md).
Boundary markers: None identified; the agent is instructed to read the full context and edit tapes directly.
Capability inventory: Uses the Bash tool to execute vhs, ffmpeg, and npm commands; uses Edit and Write tools to modify tape scripts (SKILL.md).
Sanitization: No specific sanitization or validation of user-provided command sequences for the tapes is mentioned.

record-demo