The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as job descriptions (JDs), user resumes, and search results from various career platforms (e.g., LinkedIn, Glassdoor, local job boards). This provides a surface for indirect prompt injection if the external content contains hidden or malicious instructions.
Ingestion points: Untrusted data enters the agent context through user-provided resumes and JDs, as well as job-related content retrieved via the WebSearch tool and external platforms mentioned in SKILL.md and references/interview-playbook.md.
Boundary markers: The skill utilizes a 'Reliability Playbook' (references/reliability-playbook.md) which instructs the agent to distinguish between verified facts and low-confidence data using markers like "待确认" (to be confirmed) and "低置信" (low confidence).
Capability inventory: The skill is capable of writing Markdown files to the local file system, creating Microsoft Word documents using the python-docx library, and executing local scripts such as scripts/generate_resume.py and scripts/quality_check.py.
Sanitization: The scripts/generate_resume.py script performs basic sanitization by removing specific internal labels (e.g., "迁移句:") from final outputs, but it does not implement broad sanitization against instructional injections within the source data.

get-job