writing-skills

Warn

Audited by Gen Agent Trust Hub on Sep 30, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill provides a Node.js script (render-graphs.js) that performs dynamic execution by invoking the system dot command (Graphviz) using child_process.execFileSync. The script extracts data from markdown code blocks and passes it directly to the utility's standard input. Processing external or untrusted file content through system utilities can lead to command-related vulnerabilities if the input is malformed or malicious.
  • [COMMAND_EXECUTION]: The skill's instructions encourage the execution of local scripts and helper tools, such as node render-graphs.js and bash scripts/tool.sh, to facilitate the skill development and visualization workflow.
  • [PROMPT_INJECTION]: The persuasion-principles.md file and related documentation provide a structured approach to overriding agent behavior through psychological influence techniques. By employing principles like "Authority" (using imperative language such as "YOU MUST") and "Commitment" (forcing explicit choices), the skill aims to eliminate an agent's ability to "rationalize" away from established rules. This represents a form of meta-instructioning designed to constrain and bypass agent internal reasoning.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process and evaluate other skills. It ingests data from markdown files, which are then analyzed for patterns, keywords, and graphviz code. This creates a surface for indirect injection if the processed files contain malicious instructions or malformed data designed to exploit the agent's context or the rendering utility during the analysis phase.
  • Ingestion points: Reads SKILL.md and referenced files to extract keywords, symptoms, and Graphviz code.
  • Boundary markers: None observed in the extraction scripts; the render-graphs.js script uses regex to find blocks without sanitizing the content.
  • Capability inventory: Contains subprocess execution (execFileSync) and file writing capabilities.
  • Sanitization: Missing sanitization for Graphviz code blocks before passing them to the system binary.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 30, 2026, 10:09 AM