The Agent Skills Directory

[SAFE]: The skill's operations are confined to local file management and data processing. It does not have access to network tools, which prevents the exfiltration of the sensitive personal and company data it processes.
[COMMAND_EXECUTION]: Shell tool access is restricted to a set of benign commands (mkdir, test, date) used exclusively for directory creation, file existence checks, and generating timestamps for draft filenames.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes research briefs that may originate from external sources.
Ingestion points: The skill reads research data from research/companies/ and user data from ground-truth.md.
Boundary markers: Not explicitly defined; data is directly interpolated into template variables.
Capability inventory: Restricted to reading and writing local files and limited shell commands.
Sanitization: There is no explicit sanitization of the research data before it is placed in the email draft.
Mitigation: The impact is low because the skill is incapable of autonomous action (sending) and requires the user to manually review the generated draft in the filesystem.

draft-outreach