The Agent Skills Directory

[SAFE]: The skill is strictly educational and provides best practices for managing long-context sessions. All examples are standard techniques for context retention and mitigation of attention decay.
[COMMAND_EXECUTION]: Includes a shell example for writing task state to a local JSON file in the .claude/state/ directory. This is a legitimate implementation of the scratchpad pattern for persisting state across conversation cycles and does not access sensitive system paths.
[PROMPT_INJECTION]: The skill uses directive language like 'CRITICAL CONTEXT' and 'PERSISTENT FACTS' in its examples. These are intended for reliability and fact preservation in long contexts rather than malicious injection or safety filter bypass.

context-reliability