The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the persistent storage of user-provided content, presenting an indirect prompt injection surface. Insights captured from chat history are stored in a knowledge base and are explicitly intended for future retrieval by other agents (e.g., the 'learnings-researcher'), potentially allowing malicious instructions to influence downstream tasks.\n
Ingestion points: Insights extracted from conversation history and manual /compound command inputs.\n
Boundary markers: The skill uses YAML and Markdown templates but lacks explicit instructions to downstream agents to ignore embedded commands in stored content fields.\n
Capability inventory: The skill has access to Bash, Write, and Grep for file management.\n
Sanitization: A blocking 'Validation Gate' (Step 5) enforces schema constraints, and Step 4 sanitizes filenames to mitigate path traversal and shell injection.\n- [COMMAND_EXECUTION]: The skill uses Bash and Grep to search directories and create files (e.g., mkdir -p and pattern searching). These operations are protected by strict input validation and metadata enums defined in schema.yaml to prevent arbitrary code execution.

compound