The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill-report.json file contains self-referential safety claims, including explicit verdicts such as "risk_level": "safe" and "safe_to_publish": true. This is a self-referential pattern (Cat 8e) used to attempt to override or influence automated security scanners and agent evaluations.\n
Ingestion points: Metadata fields within skill-report.json are part of the skill's data and are processed by the agent context.\n
Boundary markers: None present for the metadata content to prevent the agent from treating these claims as authoritative.\n
Capability inventory: Analysis of the skill instructions and examples confirms it lacks subprocess execution, network access, or file-writing tools.\n
Sanitization: No sanitization or validation of the metadata content is performed by the skill.\n- [SAFE]: The core instructions in SKILL.md and the visual examples provided in examples/devprep-architecture.md are purely text-based. No evidence of credential harvesting, remote code execution, or malicious persistence was found.

ascii-visualizer