The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted job description and resume content which could contain hidden instructions to manipulate the scoring outcome or influence the agent's behavior.
Ingestion points: The skill accepts job description text or file paths as an argument and references an external resume.yaml file (found in SKILL.md).
Boundary markers: The instructions do not specify any delimiters or safety prompts to ensure the agent ignores instructions embedded within the job descriptions or resumes.
Capability inventory: The skill is instructed to perform file-write operations to create session logs and summaries in the knowledge/sessions/ directory.
Sanitization: The skill uses company-slug and role-slug derived from user input to construct file paths without specifying sanitization requirements, which could theoretically be exploited for directory traversal if the agent does not apply its own safety filters.

score