The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill recognizes job postings and company websites as untrusted third-party data. It implements a mandatory sanitization workflow using tools/posting_guard.py to remove hidden text, invisible styling, and malicious instructions before the LLM processes the content.
[INDIRECT_PROMPT_INJECTION]: A strict trust boundary is enforced in 09-web-research.md, explicitly forbidding the agent from following URLs found within a posting's body or following directions embedded in fetched content.
[COMMAND_EXECUTION]: The workflow involves executing local Python utilities (tools/fetch_page.py, tools/robots_check.py, tools/verify_pdf.py, salary_lookup.py) and LaTeX typesetting engines (lualatex, xelatex) to automate research and document creation.
[COMMAND_EXECUTION]: Cross-skill tool invocation is used to interact with job portal APIs (e.g., bun run .agents/skills/<portal-skill>/cli/src/cli.ts) which provides a safer, authenticated alternative to general web fetching.
[DATA_EXPOSURE]: The skill manages highly sensitive Personal Identifiable Information (PII) in 01-candidate-profile.md and 02-behavioral-profile.md. This data is handled locally to tailor application materials and is protected by the aforementioned web research trust boundaries that prevent unauthorized exfiltration to untrusted domains.
[DYNAMIC_EXECUTION]: Document generation via LaTeX templates includes mandatory instructions to escape special characters (e.g., &, %, $, #) in user-provided or externally sourced strings, mitigating LaTeX injection and potential command execution vulnerabilities within the typesetting engine.
[SAFE]: Web research follows a responsible escalation order, including a mandatory check of robots.txt via tools/robots_check.py before retrying fetches with browser headers.

job-application-assistant