The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive process management and orchestration:
converge.py uses subprocess.Popen and subprocess.run to start, monitor, and manage the background generation process (12_qra.py).
It uses pgrep and os.kill (SIGTERM/SIGKILL) to stop and restart the generation pipeline during convergence cycles.
run.sh delegates tasks to other local skills (review-sparta, reality-check-sparta, dogpile, ask) by executing their respective shell scripts or Python entry points.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its core functionality:
Ingestion points: Processes 'QRA' (Question-Reasoning-Answer) records from a DuckDB database (sparta.duckdb), which are generated by an external LLM pipeline.
Boundary markers: The assessment logic in brandon_reviewer.py uses local Python checks (keyword matching, grounding scores) but lacks explicit delimiters or instructions to ignore embedded commands when data is passed to research/consultation tools.
Capability inventory: The skill can execute shell commands, manage processes via os.kill, and perform file system operations like shutil.copy2 on databases.
Sanitization: Implements framework-specific keyword validation (e.g., D3FEND, CWE terms) and grounding score thresholds to filter low-quality outputs, providing a layer of validation against malformed or generic content.
[EXTERNAL_DOWNLOADS]: run.sh performs runtime dependency management using uv pip install to ensure required libraries (typer, duckdb, httpx, etc.) are available in the local environment.

sparta-review