The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent in SKILL.md to execute shell commands (open, xdg-open, or start) to automatically open a generated HTML report. The file path for this report is constructed using company names provided by the user. While the skill includes instructions to slugify these names (replacing spaces with dashes, etc.), a failure or bypass in this sanitization process by the agent could allow a malicious user to inject additional shell commands into the execution string.
[DATA_EXFILTRATION]: The skill is designed to ingest and store highly sensitive personal and financial data, including base salary, bonuses, RSU vesting schedules, sign-on bonuses, visa status, and family circumstances. This information is saved in plain text within HTML reports on the user's desktop and in a local cache directory (offer-bank/) within the skill's folder. Although no network exfiltration to remote servers was detected, the local storage of such comprehensive personal profiles in unencrypted files represents a significant data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by processing untrusted data (offer details) and embedding it directly into an HTML report and shell command strings. Maliciously crafted offer details could attempt to escape the command string or inject cross-site scripting (XSS) payloads into the HTML report. When the user opens the report in a browser as instructed, injected scripts could execute in the context of the local file system.

offer-compare-skill