The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses local session files and prompt history to extract learnings. This data is stored locally in ~/.claude/learnings-queue.json and is not transmitted externally.
[COMMAND_EXECUTION]: Uses shell scripts for processing data and managing the queue. These operations are limited to local file manipulation and text processing using standard tools like jq and grep.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (prompts) to generate instructions for future sessions. While this presents an injection surface, the risk is mitigated by a mandatory human-in-the-loop review process where the user must approve every proposed change via the AskUserQuestion tool.
[DYNAMIC_CONTEXT_INJECTION]: Utilizes the dynamic context syntax (!command) in reflect.md and skip-reflect.md to load the current queue state into the agent's context. The executed commands (e.g., cat, jq, pwd) are safe and only read the plugin's own internal metadata or current directory path.

claude-reflect