The Agent Skills Directory

[COMMAND_EXECUTION]: The skill shells out to a local Python script convergence-check.py using JSON data passed via an echo pipe to determine if the analysis loop should continue. The script path is resolved using the internal ${CLAUDE_PLUGIN_ROOT} environment variable.
[DATA_EXPOSURE]: The skill reads and writes state and plan data to local project directories, such as .anneal/ and plans/. This is necessary for the orchestration of the loop and does not involve unauthorized access or exfiltration.

deepen-loop