The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill documentation includes commands that pipe remote shell scripts directly to bash from an untrusted domain (canifi.com), allowing for arbitrary code execution on the host system.\n
Evidence: curl -sSL https://canifi.com/skills/anytype/install.sh | bash in SKILL.md.\n
Evidence: curl -sSL https://canifi.com/install.sh | bash in SKILL.md.\n- EXTERNAL_DOWNLOADS (HIGH): The installation and setup processes rely on resources hosted on canifi.com, which is not a trusted source according to established safety guidelines.\n- CREDENTIALS_UNSAFE (HIGH): The skill explicitly suggests storing service passwords in local environment variables using a custom, unverified utility (canifi-env).\n
Evidence: canifi-env set SERVICE_PASSWORD "your-password" in SKILL.md.\n- COMMAND_EXECUTION (MEDIUM): The use of Playwright for automation provides the agent with the capability to perform actions in a browser environment, which could be exploited if directed by malicious content or combined with other vulnerabilities.\n- INDIRECT_PROMPT_INJECTION (LOW): The skill ingests data from external objects in Anytype via Playwright, creating a potential surface for indirect prompt injection.\n
Ingestion points: Anytype objects, blocks, and search results processed via Playwright.\n
Boundary markers: Absent.\n
Capability inventory: Object creation, deletion, linking, and browser navigation.\n
Sanitization: Not documented.

anytype