The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to run local utility scripts (read_file.py, run_tests.py) and standard development tools (e.g., npm, pytest) for task implementation and validation.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the task_file input, as the instructions explicitly command the agent to 'Trust the task spec' as the 'single source of truth' and avoid searching for external context. Ingestion points: The task_file path provided as an input parameter. Boundary markers: The task specification utilizes markdown sections (e.g., ## Objective, ## Tasks), but the skill lacks instructions to ignore or report potential malicious instructions embedded within these tasks. Capability inventory: The skill can read local files (including .claude/config.yaml), write code to implementation directories, and execute shell commands for testing. Sanitization: No validation or sanitization of the content within the task specification file is performed prior to execution.

execute-task