The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute external CLI utilities (gemini and opencode). The command construction pattern interpolates a [prompt] variable directly into the shell string (e.g., gemini -p "[prompt]"). This is a classic injection vector where shell metacharacters in the prompt could be used to execute unintended commands.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to check for the presence of external tools and prompt the user for installation if they are missing. While gemini is a well-known service, opencode (referenced with the opencode/grok-code model) is an unverified dependency that the agent is encouraged to fetch and install.
[REMOTE_CODE_EXECUTION]: By combining dynamic command construction with shell execution, the skill creates a surface for remote code execution. An attacker could provide a malicious prompt or create a file with a name that, when processed, executes arbitrary code on the host machine.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by reading untrusted data from the filesystem during the scouting process.
Ingestion points: The skill crawls directories and reads file paths to identify relevant files for tasks.
Boundary markers: Absent; there are no specified delimiters or instructions to ignore malicious content embedded within the files being scouted.
Capability inventory: The skill leverages the Bash tool for command execution and the Task tool for sub-agent orchestration.
Sanitization: Absent; the workflow does not include steps to sanitize or validate the content of the files or directory names before they influence subsequent agent actions.

scout-ext