The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's workflow includes processing untrusted external data such as existing résumé PDFs and LinkedIn profiles (Workflow Step 1 in SKILL.md). * Ingestion points: The skill accepts existing résumé PDFs and LinkedIn profile data from the user. * Boundary markers: While the skill instructs the agent to eyeball conversion results and diff against a fact sheet, it does not mandate the use of explicit delimiters or markers for the untrusted input. * Capability inventory: The skill can execute shell commands via subprocess.run (to typst and poppler-utils) and perform network requests via urllib.request. * Sanitization: Instructions in SKILL.md and ats.md guide the agent to verify facts and check for hidden text in the input to avoid obedience to embedded instructions.
[EXTERNAL_DOWNLOADS]: The script scripts/harvest_icons.py retrieves SVG path data from the well-known https://api.iconify.design service. This is a functional requirement to include contact and platform marks without vendoring large assets, but it does involve a network request to an external service at runtime.
[COMMAND_EXECUTION]: Several validation and generation scripts (verify.py, pick_design.py, verify_academic.py) use subprocess.run() to execute shell commands. These scripts invoke the typst binary and poppler-utils utilities (pdfinfo, pdftotext, pdftoppm) to build and analyze documents. These operations are aligned with the skill's primary purpose and are explicitly scoped in the allowed-tools frontmatter.

vitae