The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses the sensitive file path credentials/secrets/store.env to retrieve the AGENTMAIL_API_KEY, which is used locally for service authentication.
[EXTERNAL_DOWNLOADS]: Recommends the installation of external dependencies including the agentmail-cli and agentmail packages from public NPM and PyPI registries.
[PROMPT_INJECTION]: The skill processes inbound emails from untrusted sources, creating a potential surface for indirect prompt injection.
Ingestion points: Email content is retrieved in references/recipes.md using the agentmail inboxes:threads get command.
Boundary markers: The recipes employ timestamp, subject, and sender filtering to isolate specific, expected communications.
Capability inventory: The skill relies on subprocess execution of the agentmail CLI, jq, and grep.
Sanitization: Safety guidelines in SKILL.md advise extracting only specific data such as codes or links and warn against acting on unsolicited messages.

agentmail