The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically job posting URLs and company websites, to guide the agent's workflow.
Ingestion points: Step 1 ([job-url]) and Step 2 ([company-domain]) in SKILL.md utilize curl and Jina Reader to fetch external content.
Boundary markers: Absent. There are no instructions provided to the agent to differentiate between its own instructions and content fetched from external sources.
Capability inventory: The agent is equipped with browser automation tools (mcp__claude-in-chrome__*), shell command execution (curl), and WebSearch capabilities.
Sanitization: No sanitization or validation logic is specified for the fetched content before it is processed for contact extraction.
[COMMAND_EXECUTION]: The skill provides instructions for shell commands where parameters are interpolated from potentially untrusted variables.
Evidence: In SKILL.md Step 2, the command curl -sL "https://[company-domain]/team" incorporates the [company-domain] variable. If this variable, extracted from user input or other external sources, contains shell metacharacters, it could lead to unintended command execution.

find-linkedin-contacts