The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved via web research tools and email metadata searches, creating a surface for indirect prompt injection. The skill provides clear instructions to summarize content and requires explicit user confirmation for any resulting data writes to mitigate this risk. Ingestion points include web research summaries, email snippets, and user-provided meeting transcripts. Boundary markers are present in the form of instructions to differentiate interpretation from fact. The skill possesses capabilities for shell command execution and filesystem writes. Sanitization relies on mandatory human-in-the-loop verification before updates.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the official vendor package @getdex/cli from the NPM registry and configures the MCP server using tools hosted on the vendor's domain mcp.getdex.com.
[CREDENTIALS_UNSAFE]: The skill defines procedures for managing API credentials stored locally in ~/.dex/api-key and ~/.clihub/credentials.json. These instructions include security hardening steps such as using chmod 600 and umask 077 to restrict access, alongside warnings to never echo or print the secrets in chat or logs.
[COMMAND_EXECUTION]: The skill uses the dex CLI tool for relationship management operations. These commands are standard for the skill's purpose and utilize the binary provided by the vendor.

dex-skill