The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text from a brief and interpolates it into a prompt sent to a secondary AI agent via the acpx tool. This creates a potential surface for indirect prompt injection attacks where malicious instructions in the brief could influence the generated output.
Ingestion points: The scripts/spawn_drawer.py script reads text from a user-specified brief file.
Boundary markers: The interpolation prompt uses triple-dash delimiters (---) to separate the brief from the system instructions.
Capability inventory: The secondary agent is executed with the --deny-all flag and constrained to an empty temporary directory, limiting its ability to interact with the host system.
Sanitization: The skill extracts the resulting JSON specification using a dedicated extraction function that validates the structural integrity of the response.
[EXTERNAL_DOWNLOADS]: The documentation suggests installing the diagram-review skill using npx skills add Paldom/diagram-skills. This is a reference to a resource within the author's own ecosystem.
[COMMAND_EXECUTION]: The skill uses subprocess.run to call the local Graphviz dot utility for diagram layout and the acpx CLI for agent communication. These executions are intended for core functionality and are performed within a restricted context where applicable.
[DATA_EXFILTRATION]: The scripts/gemini_icon.py script transmits prompt data to Google's Generative Language API. This is an optional feature for generating custom icons and targets a well-known service.

illustration-draw