The Agent Skills Directory

CREDENTIALS_UNSAFE (MEDIUM): The skill instructs the agent to source sensitive credentials from environment variables (DFINITY_FORUM_PASSWORD) and local configuration files such as ~/.config/dfinity-forum/credentials.env. While this is a common pattern for CLI tools, accessing sensitive file paths constitutes a data exposure risk. The severity is mitigated to MEDIUM as this is necessary for the skill's primary function of authenticated forum research.
EXTERNAL_DOWNLOADS (LOW): The skill requires installing agent-browser from github.com/vercel-labs/agent-browser. Per the [TRUST-SCOPE-RULE], this is a trusted source, and the reference is classified as LOW severity.
COMMAND_EXECUTION (LOW): The documentation includes instructions to execute shell commands (e.g., npx skills add, agent-browser launch). These are functional requirements for the skill and target a trusted dependency.
INDIRECT PROMPT INJECTION (LOW): The skill is designed to ingest and summarize content from a public web forum, which is a known surface for indirect prompt injection (Category 8).
Ingestion points: Forum topic content, user profiles, and search results from forum.dfinity.org via JSON endpoints.
Boundary markers: Absent. The instructions do not specify delimiters or warnings for the agent to ignore instructions embedded in the forum data.
Capability inventory: The agent has access to agent-browser, which can navigate URLs, fill forms, and click elements, creating a significant impact if an injection is successful.
Sanitization: No explicit sanitization or validation of the ingested forum content is performed before processing.

forum-research