The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill requires users to store UniFi controller credentials, including username and password, in a plaintext file (config.json).
Evidence: SKILL.md and config.example.json instruct the user to provide and save real credentials in the local skill directory.
[PROMPT_INJECTION] (HIGH): Indirect Prompt Injection risk (Category 8). The skill ingests untrusted data from the network environment which can influence agent behavior.
Ingestion points: clients (hostnames), alerts (log messages/events), and devices (device names) commands in unifi-cli.ts.
Boundary markers: Absent. No instructions are provided to the agent to sanitize or ignore embedded instructions in the network data.
Capability inventory: The agent has shell execution capabilities via bun and network access via the controller API.
Sanitization: None documented; the agent parses and presents results directly to the user/context.
Risk: An attacker on the network could change a device's hostname to a malicious prompt injection payload, which the agent would then process and potentially obey.
[COMMAND_EXECUTION] (MEDIUM): The skill provides the agent with a list of CLI commands to execute on the host system using the bun runtime.
Evidence: SKILL.md provides explicit bun ~/.claude/skills/Unifi/unifi-cli.ts <command> templates.
[EXTERNAL_DOWNLOADS] (LOW): The skill downloads external dependencies during setup.
Evidence: package.json requires node-unifi from the npm registry. While node-unifi is a standard community package, it is an external dependency that executes during runtime.

Unifi