The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill explicitly instructs the agent environment to disable security sandboxing by setting dangerouslyDisableSandbox: true. This instruction is intended to bypass platform security boundaries so the grok tool can write session files to the user's home directory (~/.grok/sessions).
[COMMAND_EXECUTION]: The skill relies on executing shell commands, including the grok CLI, curl, and jq. It uses shell features like heredocs (cat <<'EOF'), background processes (&), and file redirection to /tmp/ and the user's home directory.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to api.fxtwitter.com using curl as a fallback mechanism to retrieve tweet data when the primary CLI method fails.
[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted URLs by interpolating them directly into a prompt string for the Grok CLI, creating a surface for potential injection attacks embedded in the URL or the content it points to.
Ingestion points: The <URL> placeholder in the grok command string located in SKILL.md.
Boundary markers: The command uses a heredoc to wrap the Grok prompt, but the URL is substituted directly within the instruction block.
Capability inventory: Shell execution (grok, curl), file system access (/tmp/, ~/.grok/), and network access.
Sanitization: There is no explicit validation or sanitization of the user-provided URL before it is passed to the shell command.

view-tweet