The Agent Skills Directory

PROMPT_INJECTION (LOW): Potential for indirect prompt injection via untrusted configuration files. The skill ingests YAML and JSON data from external mesh configurations which are then parsed and displayed to the agent context.\n
Ingestion points: lib/analyzer-wrapper.js reads user-specified files via fs.readFileSync.\n
Boundary markers: Absent in the provided scripts or prompts. The skill does not utilize delimiters or warnings to ignore embedded instructions in the configuration data.\n
Capability inventory: lib/analyzer-wrapper.js performs file reads; mesh-remediation.md indicates the skill has file modification capabilities via the remediation-engine.js component.\n
Sanitization: Absent. The skill uses js-yaml for parsing but does not provide specific sanitization of external content before interpolation into the agent's prompt.\n- COMMAND_EXECUTION (LOW): The skill documentation instructs the agent to execute local Node.js scripts for analysis, remediation, and reporting. While this is the intended functionality, it grants the agent the ability to execute code that interacts with the local filesystem.\n- DATA_EXPOSURE (LOW): The analyzer-wrapper.js script reads arbitrary files based on the path provided to the command line. An agent could potentially be manipulated into reading sensitive system files if directed to a non-configuration path.

mesh-security