The Agent Skills Directory

[SAFE]: The skill implements strong credential hygiene practices. It instructs the agent to load API keys from a .env file using specific shell commands (e.g., grep -c) that verify the key's existence without printing its value to the execution transcript. It also includes checks to ensure the .env file is ignored by version control.
[SAFE]: A comprehensive Confirm first policy is defined for all actions that involve spending money, contacting real people, or performing destructive operations (e.g., domain purchases, starting sequences, deleting mailboxes). This ensures that the agent cannot perform high-impact actions without explicit user approval.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as contact details and email/LinkedIn thread content, which are interpolated into outreach templates. While this presents an indirect prompt injection surface, the risk is mitigated by the mandatory human-in-the-loop confirmation required before any outreach is sent or sequence enrollment is finalized. 1. Ingestion points: GET {ws}/contacts and GET {ws}/threads (Salesforge Core API). 2. Boundary markers (absent): Variable interpolation (e.g., {{first_name}}) does not specify delimiters or sanitization in the prompt instructions. 3. Capability inventory: Sequence enrollment, thread replies, and mailbox management via curl network operations. 4. Sanitization (absent): No explicit sanitization or filtering of lead-provided variables is mentioned.

forge-suite