The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external job postings and company websites via WebFetch and WebSearch to generate application documents. This creates a surface for indirect prompt injection attacks where malicious instructions hidden in job postings could attempt to influence the agent's behavior.
Ingestion points: Job posting URLs and text retrieved in Step 1 (SKILL.md) and company research data fetched in 09-web-research.md.
Boundary markers: The skill includes strong, explicit instructions in 03-writing-style.md and 09-web-research.md (Trust Boundary section) to treat all fetched content as data only, never to follow directions embedded in it, and to avoid following links found within posting bodies.
Capability inventory: The agent has access to Bash (used for curl and LaTeX compilation), Write/Edit (used for document generation), and WebFetch/WebSearch (used for data gathering).
Sanitization: The skill provides comprehensive LaTeX character escaping rules in 05-cv-templates.md and 06-cover-letter-templates.md to prevent LaTeX injection attacks. It also includes Python-based HTML tag stripping in 09-web-research.md.
[COMMAND_EXECUTION]: The workflow involves executing shell commands via the Bash tool to perform web scraping with curl (including a bypass for bot detection firewalls) and to compile LaTeX documents using lualatex and xelatex (05-cv-templates.md, 09-web-research.md).
[DYNAMIC_EXECUTION]: The skill dynamically generates LaTeX source code incorporating both user profile data and external content, which is then compiled into PDF documents. The process includes specific mitigations for injection and instructions for visual and mechanical verification of the resulting PDFs (05-cv-templates.md).

job-application-assistant