The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to execute a shell command: python3 "${CLAUDE_CONFIG_DIR:-$HOME/.claude}/gauntlet/run.py" preflight <ref>. The <ref> value is sourced directly from user-provided arguments. If the agent does not strictly validate this input before passing it to the shell, it could lead to arbitrary command injection.
[DYNAMIC_EXECUTION]: The skill relies on the execution of an external Python script (run.py) located in the user's configuration directory. The output of this script is then used to dynamically configure and trigger a secondary workflow (gauntlet-run).
[INDIRECT_PROMPT_INJECTION]: The skill fetches issue descriptions from Linear and writes them verbatim to local files for further processing. This allows untrusted content from external systems to enter the agent's context and potentially influence downstream actions. 1. Ingestion points: Issue titles and descriptions retrieved via Linear MCP tools. 2. Boundary markers: Absent; instructions specify writing content "verbatim". 3. Capability inventory: Shell command execution (Python script), Git worktree management, and Workflow triggering. 4. Sanitization: No sanitization or filtering of the external issue content is described before it is processed.

gauntlet