The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.check_call in scripts/render.py and scripts/test_audit_prompt.py. These calls are used to orchestrate internal utility scripts (render_card.py and audit_prompt.py) for rendering result cards and running internal tests. The commands are constructed using sys.executable and target local file paths within the skill bundle.
[DYNAMIC_EXECUTION]: In scripts/render_card.py, the skill uses __import__('os') to check for environment variables and includes a conditional import cairosvg within a try-except block. This is a standard pattern for handling optional dependencies (the cairosvg library) for specific tasks like PNG rendering.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and optimize user-provided system prompts, which represents an attack surface for indirect prompt injection.
Ingestion points: User-specified files are read by scripts/audit_prompt.py and scripts/scaffold_promptfoo.py.
Boundary markers: The scaffold_promptfoo.py script appends a "User task:" marker to the ingested prompt content during preparation.
Capability inventory: The skill can execute internal scripts via subprocess and recommends the use of external tools like promptfoo for behavioral testing.
Sanitization: The audit logic in scripts/audit_prompt.py uses regex to identify prompt engineering patterns but does not explicitly sanitize the content for malicious instructions.
[EXTERNAL_DOWNLOADS]: The skill's documentation (SKILL.md and references/SOURCES.md) recommends the use of promptfoo and references public repositories from well-known organizations like Anthropic and Sentry as standard primitives for the prompt optimization workflow.

system-prompt-doctor