The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to interact with the local file system, specifically for managing files within the $HOME/.claude/instincts/ directory structure. This includes listing files, reading logs, and moving files to an archive.
[DYNAMIC_EXECUTION]: The skill relies on local Python libraries (confidence, similarity) stored in a specific library path to perform its logic. It constructs and executes Python snippets via python3 -c using values extracted from file frontmatter. This is a functional requirement for the skill's lifecycle management tasks.
[INDIRECT_PROMPT_INJECTION]: The skill processes event logs to adjust the confidence scores of its behavioral rules. While these logs could potentially contain data influenced by external interactions, the impact is limited to internal metadata updates within the agent's restricted environment.
Ingestion points: Event log files (.ndjson) in the $HOME/.claude/instincts/events directory.
Boundary markers: None; the skill assumes a specific log format for parsing.
Capability inventory: Uses Write, Bash, and Glob tools to modify and move local behavioral files.
Sanitization: No explicit sanitization of log contents is described in the evaluation logic.

learn-eval