The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied email drafts, recipient lists, and DNS records as part of its core auditing function. It also suggests browsing external websites to verify provider guidelines and domain authentication status. This ingestion of untrusted data from external sources creates a surface for indirect prompt injection, where malicious instructions could be embedded in the processed data to influence the agent's behavior.
Ingestion points: SKILL.md (Workflow steps 3, 4, and 5) and evals/evals.json indicate the agent ingests draft text and DNS output.
Boundary markers: The "Ethics boundary" and "Quality gate" sections in SKILL.md define strict behavioral constraints, though they do not use explicit text delimiters for user input.
Capability inventory: The skill utilizes dig commands for DNS verification and web browsing for documentation review (SKILL.md, references/authentication-and-setup.md).
Sanitization: No specific sanitization or filtering logic is defined for the input data; the skill relies on the agent's task-specific critique logic.
[COMMAND_EXECUTION]: The skill provides templates for shell commands using the dig utility (dig TXT <domain> +short) to perform verification of SPF, DKIM, and DMARC records. These commands are standard administrative tools for the skill's intended purpose of deliverability auditing and do not represent unauthorized or malicious system access.

cold-email-deliverability