The Agent Skills Directory

[Data Exposure] (HIGH): The common.sh script employs a high-risk method for loading environment variables. It calculates a path five levels above its installation directory—which resolves to the user's home directory (~) if installed as instructed—and attempts to load a .env file from that location. Crucially, the script exports every variable found in the file into the current process environment. This can lead to the unintended exposure of highly sensitive secrets (e.g., AWS keys, private tokens) that may be stored in a general-purpose .env file in the home directory.
[Indirect Prompt Injection] (LOW): The skill ingests and processes untrusted data from the Unsplash API, creating a vulnerability to indirect prompt injection.
Ingestion points: scripts/search.sh and scripts/random.sh receive data from api.unsplash.com.
Boundary markers: Absent; the skill does not use delimiters or instruct the agent to ignore instructions within the API response.
Capability inventory: The skill uses curl for networking and jq for data extraction, providing formatted strings (like photo descriptions and attribution HTML) that the agent may interpret as commands.
Sanitization: Absent; data is passed directly from the API to the agent without filtering or escaping.
[External Downloads] (LOW): The skill makes network requests to api.unsplash.com. While this is the intended purpose of the skill, it involves communication with a non-whitelisted external domain.

unsplash