The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and modify a local configuration file (~/.pua/config.json). This creates an attack surface where instructions embedded in the configuration data could influence the agent's behavior.
Ingestion points: The agent is directed to load configuration from ~/.pua/config.json in SKILL.md.
Boundary markers: No delimiters or instructions to ignore embedded commands are present.
Capability inventory: The skill has capabilities to read and write to the filesystem to manage the configuration.
Sanitization: No sanitization or validation of the configuration content is described.
[DATA_EXFILTRATION]: The skill instructions require the agent to perform 'platform telemetry' and 'KPI' collection. This directs the agent to gather data about the host platform and its performance, which could involve sensitive metadata or environment details. While no network destination is specified in this file, the data collection behavior is a prerequisite for exfiltration.

pua-pro