The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes arbitrary shell commands using subprocess.run(shell=True) in the execute_atomic_manual and run_cleanup functions. These commands are parsed from YAML test definitions.
[REMOTE_CODE_EXECUTION]: The skill workflow fetches test definitions from the redcanaryco/atomic-red-team GitHub repository and executes commands defined within them locally.
[EXTERNAL_DOWNLOADS]: Documentation in SKILL.md and references/api-reference.md directs users to download the atomic-operator library and clone the atomic-red-team repository.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of external test definitions. Ingestion points: scripts/agent.py (load_atomic_tests function). Boundary markers: Absent. Capability inventory: subprocess.run(shell=True) in scripts/agent.py. Sanitization: Absent; the script performs string replacement for arguments without escaping shell characters.

performing-threat-emulation-with-atomic-red-team