The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The documentation describes a system that ingests untrusted user prompts and tool inputs to inject context or block operations. • Ingestion points: UserPromptSubmit and PreToolUse hooks (described in HOOK_MECHANISMS.md). • Boundary markers: Visual delimiters (e.g., box drawing characters) are used in output examples. • Capability inventory: Can inject system messages and block tool execution via exit codes. • Sanitization: No explicit sanitization of ingested content is described in the documentation.
[No Code] (SAFE): The skill contains only Markdown documentation. No executable scripts (.ts, .sh) or JSON configuration files (skill-rules.json) were included in the analyzed set.

skill-developer