The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates reading and responding to emails via the EtherMail service. Content from these emails is provided by external third parties and could contain malicious instructions designed to manipulate the agent's logic once processed.
Ingestion points: Browser-based inbox access as described in SKILL.md.
Boundary markers: No specific delimiters or warnings to ignore instructions within email content are implemented.
Capability inventory: Includes shell command execution (via node scripts), browser automation, and wallet signing capabilities via the walletconnect-agent skill.
Sanitization: No sanitization or filtering of email body content is specified.
[DYNAMIC_EXECUTION]: The helper script scripts/extract-wc-uri.js utilizes Puppeteer's page.evaluate() function to run custom JavaScript within the browser context to navigate the EtherMail Shadow DOM and extract connection strings.
[CREDENTIALS_UNSAFE]: The documentation provides instructions to export PRIVATE_KEY="0x..." in the shell environment. While the skill includes a warning about committing keys, storing high-value secrets like Ethereum private keys in environment variables makes them vulnerable to exposure via process monitoring or shell history files.
[METADATA_POISONING]: The primary Telegram link and several references in SKILL.md include an affiliate tracking parameter (afid_6986e9a5c5a97b905a78c390), which is used to attribute user registrations to the skill author.

ethermail