The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data provided by the user (such as resumes, company names, and job descriptions) without using boundary markers or sanitization.
Ingestion points: Data enters the agent context through user input fields for offer details, salary structures, and optional resumes shared from other skills as described in the 'Information Collection' section of SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard potential instructions embedded within the user-provided offer data or resumes.
Capability inventory: The skill generates detailed text-based comparisons, calculates financial metrics (Annual Package, Hourly Wage), and produces HTML reports based on the design specification in references/report-spec.md.
Sanitization: The instructions do not define any sanitization, escaping, or validation protocols for the external content before it is interpolated into analysis prompts or HTML generation blocks.
[DATA_EXFILTRATION]: While no explicit exfiltration to external URLs was detected, the skill is designed to collect highly sensitive personal information, including detailed salary components, company names, equity/stock details, and professional resumes. This data collection increases the risk of exfiltration if the conversation context is shared or if the agent environment is compromised.

offer-decision