The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The sibling_setup.sh script includes logic to clone the vendor's official claude-skills repository from GitHub if the scripts are not already present in the environment.- [COMMAND_EXECUTION]: The harness relies heavily on system command execution via Python's subprocess module to perform git operations (worktree management, merging, committing), execute project-specific test runners, and manage subagent life cycles.- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests failure reasons and stage labels from external test runner output, which are then interpolated into prompts for subagents. This risk is managed through a normalization process in workqueue.py that masks specific identifiers, hex values, and paths to create stable failure signatures.

swarming-failing-tests