The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute duckdb queries. It dynamically constructs commands to search through local session logs based on user-provided keywords and project paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from past session logs which could potentially contain malicious instructions intended to influence the agent's behavior during internalization. 1. Ingestion points: Data is read from session logs in the $HOME/.claude/projects/ directory. 2. Boundary markers: None explicitly defined to separate log content from instructions. 3. Capability inventory: The skill has access to the Bash tool to execute filesystem queries. 4. Sanitization: The instructions direct the agent to internalize findings rather than outputting them directly, which acts as a basic safeguard, though no formal sanitization of the log content is performed.

read-memories