The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, which creates a surface for potential indirect prompt injection attacks.
Ingestion Points: In SKILL.md, the instructions explicitly account for processing user-provided files, pasted content, screenshots, tables, and briefs.
Boundary Markers: The skill attempts to mitigate risks by conditionally loading a specific reference file (../eva-shared/references/shared/06_external-material-safety_外部材料安全边界.md) when external materials are provided.
Capability Inventory: Based on the provided files, the skill's capabilities are limited to text analysis, multi-disciplinary divergence, and routing to other internal modules. It does not contain scripts for file-system modification, remote code execution, or network exfiltration.
Sanitization: The skill includes logical constraints in SKILL.md and references/lens/03_evidence_证据与出口边界.md that prevent the model from simulating figures, fabricating citations, or making clinical diagnoses, which serves as a defensive layer against manipulated inputs.
[NO_CODE]: The skill consists entirely of Markdown instructions and YAML configuration. No executable scripts (e.g., Python, JavaScript, Shell) were detected, reducing the risk of direct technical exploits.

eva-lens