The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which creates a surface for indirect prompt injection attacks.
Ingestion points: The skill reads and parses content from user-specified directories and files in various formats, including PDF, DOCX, HTML, Markdown, and YAML (SKILL.md, Phase 1 & 2).
Boundary markers: There are no explicit instructions to use delimiters or ignore instructions embedded within the source documents.
Capability inventory: The skill has the capability to write to the file system (resume.yaml) and execute shell commands (resume-builder verify) (SKILL.md, Phase 5).
Sanitization: While the skill includes 'Anti-Fabrication Rules' and a user review phase (Phase 4), it lacks explicit sanitization or filtering to prevent malicious instructions hidden in resume documents from influencing the agent's behavior.
[COMMAND_EXECUTION]: The skill performs local command execution as part of its verification process.
Evidence: In Phase 5, the skill is instructed to run the resume-builder verify command to validate the newly written resume.yaml file.

import