The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform administrative tasks within the project environment. This includes running a local Python kernel (context/kernel.py) for lock management, using grep to search for existing facts, and employing git stash and git diff to implement a safety layer for file modifications.
[DATA_EXFILTRATION]: File access is restricted to the project's memory and context directories. There are no network-enabled commands or operations that attempt to send data to external servers.
[PROMPT_INJECTION]: The skill instructions are purely functional and do not contain attempts to override agent safety guidelines. It includes 'The Dementia Defense' as a safeguard to prevent the agent from blindly adding conflicting information to the long-term memory.
[REMOTE_CODE_EXECUTION]: The skill does not perform any remote downloads or execute code from untrusted external sources. It relies on standard system libraries and local scripts defined within the Agentic OS framework.

os-memory-manager