The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute system commands such as open (macOS), xdg-open (Linux), or start (Windows) to automatically launch the generated HTML report in the user's browser upon completion of the analysis flow.- [DATA_EXFILTRATION]: The skill performs file system operations, specifically writing generated report files to the ~/Desktop/Claude skills/ directory and reading from or writing to a local jd-bank/ folder for analysis caching. This involves interaction with the user's local filesystem and desktop environment.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically job descriptions provided via links or text pastes, which presents a vulnerability surface for indirect prompt injection.
Ingestion points: External job description text and URL links (as seen in SKILL.md Step 1).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when interpolating external job description data into processing prompts.
Capability inventory: File system write access to the user's desktop and automated command execution for opening local files.
Sanitization: No specific content sanitization or instruction-filtering mechanisms are mentioned for the external JD data.- [EXTERNAL_DOWNLOADS]: The skill's HTML template (examples/offer-strategy-template.html) references external assets from well-known services, specifically Google Fonts (fonts.googleapis.com) to ensure consistent document rendering. These are considered safe well-known resources.- [METADATA_POISONING]: The instructions mandate the inclusion of specific author branding, social media links, and footer metadata in all generated reports, ensuring persistent attribution to the author 'yanliudesign' but restricting user control over the final output content.

job-description-skill