The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted content from external Resource files in Step 1. The agent processes this content to surface ideas and terms which are then used as arguments (e.g., --reason, --query) for the idea-db CLI tool in Steps 5 and 9. This creates a surface for indirect prompt injection where a malicious resource could attempt to manipulate database commands. The skill provides logical constraints for the agent to reformat input, but lacks strict technical sanitization.\n
Ingestion points: Resource files read in Step 1.\n
Boundary markers: No explicit delimiters or instructions for the agent to ignore malicious instructions within resource files are provided.\n
Capability inventory: Shell command execution via the idea-db CLI tool.\n
Sanitization: No specific programmatic sanitization or escaping is required for content derived from resources before its use in command arguments.\n- [COMMAND_EXECUTION]: The skill requires the use of the idea-db CLI tool to manage a local SQLite database (.slipbox/idea.db). It executes commands to query existing terms, check for duplicates, and add new entries. This usage is consistent with the skill's purpose of slipbox management and interacts with local files scoped to the .slipbox directory.

surface-ideas