The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script scripts/kb_loader.py to index topics and retrieve structured metadata about knowledge bases.
[DATA_EXPOSURE]: The helper script accesses the global configuration file at ~/.claude/knowledge-base/config.json and explores the .claude/agents/ directory to identify available knowledge sources and agent definitions. This access is consistent with the skill's intended purpose of knowledge base management.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by processing untrusted Markdown content across multiple tiers (Frontmatter, Table of Contents, and Full Read) without sanitization.
Ingestion points: Markdown files identified during discovery and the knowledge base configuration file ~/.claude/knowledge-base/config.json.
Boundary markers: None. The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the knowledge base content.
Capability inventory: Execution of local scripts (kb_loader.py), file reading via agent tools, and directory traversal.
Sanitization: None. The content is extracted and presented to the agent context without escaping or validation.

kb-find