The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation and examples reference the use of sensitive environment variables such as OPENAI_API_KEY, TELEGRAM_BOT_TOKEN, SOLANA_PRIVATE_KEY, and POSTGRES_URL.
Evidence: Found in examples/telegram-bot/README.md, examples/multi-agent/README.md, and SKILL.md.
Mitigation: The skill explicitly warns users to 'Never commit .env files' and to use .env.example as a template.
[COMMAND_EXECUTION]: The installation and setup guides instruct users to execute shell commands for package management and process monitoring.
Evidence: bun i -g @elizaos/cli and pm2 start in examples/telegram-bot/README.md and SKILL.md.
[DATA_EXFILTRATION]: The custom plugin example demonstrates fetching market data from an external API.
Evidence: fetchTokenData in examples/custom-plugin/README.md makes requests to api.coingecko.com.
Analysis: This is a well-known service and the operation is consistent with the skill's stated purpose of building a token info plugin.
[INDIRECT_PROMPT_INJECTION]: The framework is designed to ingest and process data from external, untrusted sources which could contain malicious instructions.
Ingestion points: Telegram groups/DMs, Discord channels, and Twitter mentions as described in SKILL.md and client-specific READMEs.
Boundary markers: The provided character templates (e.g., templates/eliza-character.json) lack explicit boundary markers or 'ignore' instructions for embedded data, relying instead on high-level style guidelines.
Capability inventory: The framework includes the @elizaos/plugin-solana, which allows agents to perform on-chain actions like SEND_TOKEN and SWAP_TOKEN (SKILL.md).
Sanitization: The provided code examples for custom actions and providers do not show explicit sanitization or filtering of user-provided text before processing.

eliza